MLATE V3: batched optimiser, all model families, OpenRouter protocol generation

.gitattributes

@@ -1,37 +1,3 @@
-*.7z filter=lfs diff=lfs merge=lfs -text
-*.arrow filter=lfs diff=lfs merge=lfs -text
-*.bin filter=lfs diff=lfs merge=lfs -text
-*.bz2 filter=lfs diff=lfs merge=lfs -text
-*.ckpt filter=lfs diff=lfs merge=lfs -text
-*.ftz filter=lfs diff=lfs merge=lfs -text
-*.gz filter=lfs diff=lfs merge=lfs -text
-*.h5 filter=lfs diff=lfs merge=lfs -text
-*.joblib filter=lfs diff=lfs merge=lfs -text
-*.lfs.* filter=lfs diff=lfs merge=lfs -text
-*.mlmodel filter=lfs diff=lfs merge=lfs -text
-*.model filter=lfs diff=lfs merge=lfs -text
-*.msgpack filter=lfs diff=lfs merge=lfs -text
-*.npy filter=lfs diff=lfs merge=lfs -text
-*.npz filter=lfs diff=lfs merge=lfs -text
-*.onnx filter=lfs diff=lfs merge=lfs -text
-*.ot filter=lfs diff=lfs merge=lfs -text
-*.parquet filter=lfs diff=lfs merge=lfs -text
-*.pb filter=lfs diff=lfs merge=lfs -text
-*.pickle filter=lfs diff=lfs merge=lfs -text
-*.pkl filter=lfs diff=lfs merge=lfs -text
-*.pt filter=lfs diff=lfs merge=lfs -text
-*.pth filter=lfs diff=lfs merge=lfs -text
-*.rar filter=lfs diff=lfs merge=lfs -text
-*.safetensors filter=lfs diff=lfs merge=lfs -text
-saved_model/**/* filter=lfs diff=lfs merge=lfs -text
-*.tar.* filter=lfs diff=lfs merge=lfs -text
-*.tar filter=lfs diff=lfs merge=lfs -text
-*.tflite filter=lfs diff=lfs merge=lfs -text
-*.tgz filter=lfs diff=lfs merge=lfs -text
-*.wasm filter=lfs diff=lfs merge=lfs -text
-*.xz filter=lfs diff=lfs merge=lfs -text
-*.zip filter=lfs diff=lfs merge=lfs -text
-*.zst filter=lfs diff=lfs merge=lfs -text
-*tfevents* filter=lfs diff=lfs merge=lfs -text
-*.cbm filter=lfs diff=lfs merge=lfs -text
-Dataset.xlsx filter=lfs diff=lfs merge=lfs -text
+*.pkl filter=lfs diff=lfs merge=lfs -text
+*.pt filter=lfs diff=lfs merge=lfs -text
+*.parquet filter=lfs diff=lfs merge=lfs -text

README.md

@@ -1,127 +1,75 @@
----
-title: MLATE
-emoji: 🥼
-colorFrom: red
-colorTo: indigo
-sdk: streamlit
-sdk_version: 1.44.1
-python_version: "3.12"
-app_file: app.py
-pinned: false
-license: cc-by-nc-4.0
-short_description: ML Applications in Tissue Engineering
----
-
-
-# 👬 MLATE V3: Multi-Tissue Scaffold Prediction Platform
-
-**MLATE V3** is a fully integrated, machine learning-powered platform for predicting, optimizing, and generating detailed fabrication procedures for 3D-(bio)printed scaffolds in tissue engineering. This app enables researchers to input a wide range of biomaterials, cell lines, and printing parameters, and receive optimized scaffold compositions along with step-by-step printing instructions generated via Google Gemini.
-
-> 📄 *Rafieyan et al. (preprint, 2025). MLATE V3: A fully integrated Multi-Tissue, machine learning platform for prediction, optimization and generating procedures for fabricating 3D-(bio)printing scaffolds for tissue engineering*
-
----
-
-## 🚀 Features
-
-- 🔬 Predict scaffold quality based on printability and cell response
-- 🧪 Optimize biomaterial concentrations, cell densities, and printing parameters using Optuna
-- 🧠 Powered by two fine-tuned **CatBoostClassifier** models
-- 📋 Automatically generates fabrication protocols with Gemini API
-- 🔐 Enforces safe defaults and intelligent UI input validation
-- 🧱 Uses a real-world, curated dataset of **2847 samples** across **multiple tissues and cell lines**
-
----
-
-## 📂 Dataset
-
-This project includes a publicly available dataset (`Dataset.xlsx`) containing:
-- 123 biomaterials
-- 175 cell lines
-- 7 printing parameters
-- Scaffold performance labels
-
-The dataset is available in the [Files and Versions](https://huggingface.co/spaces/your-username/your-space-name/blob/main/Dataset.xlsx) tab of this Space.
-
-You may also optionally add this to Hugging Face Datasets for broader access.
-
----
-
-## ⚙️ How It Works
-
-1. **Input**: User selects biomaterials, cell line, and printing parameters with min/max/step values
-2. **Optimization**: Optuna runs 50 trials to maximize predicted scaffold quality (WSSQ)
-3. **Prediction**:
-    - Two CatBoost models are used to predict:
-        - `Printability` (3-class)
-        - `Cell Response` (5-class)
-    - Probabilistic predictions are mapped to expected scores
-4. **Scaffold Quality**: A weighted combination of printability and cell response
-5. **Procedure Generation**: A Gemini API prompt generates custom step-by-step fabrication instructions
-
----
-
-## 💻 Running Locally
-
-Clone the repo and install dependencies:
-
-```bash
-git clone https://huggingface.co/spaces/your-username/MLATE-V3
-cd MLATE-V3
-
-# Create virtual environment (optional)
-python -m venv venv
-source venv/bin/activate  # or venv\Scripts\activate on Windows
-
-# Install dependencies
-pip install -r requirements.txt
-
-# Set your Gemini API key
-export GEMINI_API_KEY=your_key_here  # or set in .env
-
-# Run the app
-streamlit run app.py
-```
-
----
-
-## 📜 License
-
-This project is licensed under the **Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NC 4.0)**.
-
-You are free to:
-- Share and adapt the code
-- Use the dataset for academic research
-
-But:
-- **Commercial use is prohibited**
-- **Citation is required** (see below)
-
----
-
-## 📚 Citation
-
-If you use MLATE V3 or its dataset in your research, please cite:
-
-> Rafieyan et al. (preprint, 2025).  
-> *MLATE V3: A fully integrated Multi-Tissue, machine learning platform for prediction, optimization and generating procedures for fabricating 3D-(bio)printing scaffolds for tissue engineering*  
-> *(Preprint link to be added after publication)*
-
-BibTeX:
-```bibtex
-@article{rafieyan2025mlate,
-  author  = {Rafieyan, Saeed and others},
-  title   = {MLATE V3: A fully integrated Multi-Tissue, machine learning platform for prediction, optimization and generating procedures for fabricating 3D-(bio)printing scaffolds for tissue engineering},
-  journal = {Preprint},
-  year    = {2025}
-}
-```
-
----
-
-## ⚠️ Disclaimer
-
-This tool is intended for **research and academic use only**. While we strive for accuracy, the predictions and fabrication procedures are generated using machine learning and language models and may contain errors or inconsistencies. The authors are **not responsible for any unintended consequences** arising from use of this tool in experimental or clinical settings.
-
----
-
-Developed by [Saeed Rafieyan](https://sraf.ir)
+---
+title: MLATE V3
+emoji: 🧬
+colorFrom: blue
+colorTo: green
+sdk: streamlit
+sdk_version: 1.58.0
+app_file: app.py
+pinned: false
+license: mit
+short_description: Optimises 3D-printed and bioprinted scaffolds
+---
+
+# MLATE V3
+
+Predicts **printability** and **cell response** for 3D-printed and bioprinted
+scaffolds, searches the formulation space for the composition and printing
+conditions that maximise a combined quality score, and drafts a bench protocol
+for the result.
+
+Models were trained on 2,646 scaffold records extracted from the literature.
+Predictions are decision support: they narrow the experimental search space and
+do not replace experimental validation.
+
+## Using it
+
+1. Enter a range for each biomaterial you can work with, and for the printing
+   parameters your equipment allows. The optimiser searches inside those ranges.
+2. Choose a cell line, or `NoCellCultured` for acellular printing.
+3. Set the cell-response weight. Printability takes the remainder.
+4. Run the optimisation.
+5. Optionally generate a fabrication protocol. This step needs a free
+   [OpenRouter](https://openrouter.ai) key; nothing else does.
+
+## What the score means
+
+WSSQ combines the two predicted outcomes through two conjunctive means, so a
+scaffold cannot score well by excelling at one objective and failing the other.
+Acellular formulations are scored on printability alone rather than penalised
+for a biological outcome that does not apply to them.
+
+## Models
+
+All three model families are offered for each target, ranked by weighted F1 on
+the held-out test partition and refitted on the complete dataset. On a GPU host
+the menu opens on the best model overall; on a CPU host it opens on the best
+conventional classifier, because the in-context foundation models re-read all
+2,646 training records on every pass and turn a search of seconds into one of
+minutes. They remain selectable, with their cost stated beside the menu.
+
+The random-split artefacts are served, this being the interpolation regime in
+which the tool is used: adjusting a concentration, substituting a cell line, or
+moving a pressure within observed ranges. The study-grouped models are the
+conservative estimate for an unseen laboratory and are reported in the paper.
+
+## Deploying your own copy
+
+From a checkout of the repository:
+
+```
+python 06_webapp/export_deployment.py    # fit and export the artefacts
+python 06_webapp/build_app_data.py       # vocabularies, tables, corpus subset
+python 06_webapp/deploy_to_hf.py         # stage and check, no upload
+python 06_webapp/deploy_to_hf.py --push  # upload, after `hf auth login`
+```
+
+`deploy_to_hf.py` assembles a self-contained tree under `deploy/space/`,
+verifies that it imports and finds its models, and uploads it. Model artefacts
+are not in this directory by default; they are added by that script.
+
+## Citation
+
+Rafieyan *et al.*, *MLATE V3: An Open-Source Cross-Tissue AI Framework for
+Data-Driven Optimization of 3D-Printed and Bioprinted Scaffolds*.
+Code and dataset: https://github.com/saeedrafieyan/mlate

app.py

@@ -1,1012 +1,633 @@
-import os
-from pathlib import Path
-import streamlit as st
-import pandas as pd
-import joblib
-import optuna
-import numpy as np
-
-from google import genai
-from google.genai import types
-
-from biomaterials import BIOMATERIAL_OPTIONS
-from cell_lines import CELL_LINE_OPTIONS
-
-_original_number_input = st.number_input
-
-def safe_number_input(label, **kwargs):
-    """
-    Clamp `value` into [min_value, max_value] and warn if we had to adjust.
-    Then call the real st.number_input with the clamped default.
-    """
-    min_value = kwargs.get("min_value", float("-inf"))
-    max_value = kwargs.get("max_value", float("inf"))
-    value     = kwargs.get("value", min_value)
-    clamped = min(max(value, min_value), max_value)
-    if clamped != value:
-        st.warning(
-            f"⚠️ Default for “{label}” ({value}) was outside "
-            f"[{min_value}, {max_value}]; using {clamped} instead."
-        )
-    kwargs["value"] = clamped
-    return _original_number_input(label, **kwargs)
-
-st.number_input = safe_number_input
-
-APP_DIR = Path(__file__).resolve().parent
-MODEL_ROOT = APP_DIR / "models"
-PREPROCESSOR_DIR = MODEL_ROOT / "preprocessors"
-
-MODEL_TASKS = {
-    "printability": {
-        "folder": MODEL_ROOT / "printability",
-        "prefix": "Printability_",
-        "label_encoder": "label_encoder_printability.pkl",
-    },
-    "cell_response": {
-        "folder": MODEL_ROOT / "cell response",
-        "prefix": "Cell_Response_",
-        "label_encoder": "label_encoder_cell_response.pkl",
-    },
-}
-
-DL_MODEL_CONFIGS = {
-    ("printability", "ResNet"): {"n_layers": 6, "hidden_dim": 302, "dropout": 0.190106, "activation_func": "relu"},
-    ("printability", "MLP"): {"n_layers": 3, "hidden_dim": 367, "dropout": 0.169472, "activation_func": "tanh"},
-    ("printability", "1D_CNN"): {"n_layers": 6, "hidden_dim": 287, "dropout": 0.233072, "activation_func": "relu"},
-    ("printability", "FT_Transformer"): {"n_layers": 3, "hidden_dim": 437, "dropout": 0.324095, "activation_func": "GELU"},
-    ("printability", "TabNet_Lite"): {"n_layers": 4, "hidden_dim": 283, "dropout": 0.309445, "activation_func": "relu"},
-    ("printability", "NODE_Lite"): {"n_layers": 5, "hidden_dim": 289, "dropout": 0.145481, "activation_func": "SELU"},
-    ("cell_response", "ResNet"): {"n_layers": 4, "hidden_dim": 269, "dropout": 0.229224, "activation_func": "tanh"},
-    ("cell_response", "MLP"): {"n_layers": 5, "hidden_dim": 238, "dropout": 0.256294, "activation_func": "ELU"},
-    ("cell_response", "1D_CNN"): {"n_layers": 6, "hidden_dim": 134, "dropout": 0.158794, "activation_func": "SiLU"},
-    ("cell_response", "FT_Transformer"): {"n_layers": 5, "hidden_dim": 395, "dropout": 0.185155, "activation_func": "SiLU"},
-    ("cell_response", "TabNet_Lite"): {"n_layers": 5, "hidden_dim": 342, "dropout": 0.103486, "activation_func": "SiLU"},
-    ("cell_response", "NODE_Lite"): {"n_layers": 2, "hidden_dim": 127, "dropout": 0.328308, "activation_func": "SiLU"},
-}
-
-MODEL_RANKINGS = {
-    "printability": [
-        "HistGradientBoosting", "TabPFN 2.6", "Bagging", "GradientBoosting", "TabICL v2",
-        "XGBoost", "LightGBM", "KNeighbors", "LabelPropagation", "ExtraTrees", "MLP",
-        "LabelSpreading", "TabNet Lite", "MLP DL", "ResNet", "1D CNN", "DecisionTree",
-        "NODE Lite", "LinearSVC", "LDA", "PassiveAggressive", "CalibratedClassifierCV",
-        "LogisticRegression", "RidgeClassifier", "FT Transformer", "Perceptron",
-        "MultinomialNB", "RadiusNeighbors", "ComplementNB", "NuSVC", "AdaBoost",
-        "SGD", "BernoulliNB", "GaussianNB", "QDA", "RandomForest", "ExtraTreeClassifier",
-        "DummyClassifier",
-    ],
-    "cell_response": [
-        "HistGradientBoosting", "TabICL v2", "TabPFN 2.6", "XGBoost", "Bagging",
-        "KNeighbors", "LabelPropagation", "LabelSpreading", "LightGBM", "FT Transformer",
-        "CalibratedClassifierCV", "ResNet", "NODE Lite", "1D CNN", "MLP DL", "LDA",
-        "NuSVC", "MultinomialNB", "LogisticRegression", "LinearSVC", "Perceptron",
-        "QDA", "RidgeClassifier", "ExtraTrees", "ExtraTreeClassifier", "AdaBoost",
-        "DecisionTree", "RandomForest", "GradientBoosting", "SGD", "MLP",
-        "PassiveAggressive", "ComplementNB", "TabNet Lite", "BernoulliNB",
-        "RadiusNeighbors", "DummyClassifier", "GaussianNB",
-    ],
-}
-
-PERFORMANCE_GUIDE = {
-    "printability": """
-| Rank | Model | Framework | Accuracy | F1 | AUC | MCC |
-|---:|---|---|---:|---:|---:|---:|
-| 1 | HistGradientBoosting | Machine Learning | 0.80 | 0.80 | 0.94 | 0.69 |
-| 2 | TabPFN 2.6 | Deep Learning / Transformer | 0.80 | 0.80 | 0.94 | 0.69 |
-| 3 | Bagging | Machine Learning | 0.79 | 0.79 | 0.93 | 0.67 |
-| 4 | GradientBoosting | Machine Learning | 0.79 | 0.79 | 0.93 | 0.66 |
-| 5 | TabICL v2 | Deep Learning / Transformer | 0.79 | 0.79 | 0.94 | 0.68 |
-| 6 | XGBoost | Machine Learning | 0.78 | 0.78 | 0.93 | 0.66 |
-| 7 | LightGBM | Machine Learning | 0.77 | 0.77 | 0.93 | 0.64 |
-| 8 | KNeighbors | Machine Learning | 0.77 | 0.77 | 0.90 | 0.64 |
-| 9 | LabelPropagation | Machine Learning | 0.77 | 0.77 | 0.82 | 0.64 |
-| 10 | ExtraTrees | Machine Learning | 0.77 | 0.77 | 0.92 | 0.63 |
-""",
-    "cell_response": """
-| Rank | Model | Framework | Accuracy | F1 | AUC | MCC |
-|---:|---|---|---:|---:|---:|---:|
-| 1 | HistGradientBoosting | Machine Learning | 0.81 | 0.81 | 0.96 | 0.67 |
-| 2 | TabICL v2 | Deep Learning / Transformer | 0.79 | 0.79 | 0.97 | 0.65 |
-| 3 | TabPFN 2.6 | Deep Learning / Transformer | 0.78 | 0.78 | 0.96 | 0.63 |
-| 4 | XGBoost | Machine Learning | 0.79 | 0.77 | 0.96 | 0.65 |
-| 5 | Bagging | Machine Learning | 0.77 | 0.77 | 0.95 | 0.62 |
-| 6 | KNeighbors | Machine Learning | 0.77 | 0.77 | 0.93 | 0.61 |
-| 7 | LabelPropagation | Machine Learning | 0.77 | 0.77 | 0.93 | 0.61 |
-| 8 | LabelSpreading | Machine Learning | 0.78 | 0.76 | 0.93 | 0.61 |
-| 9 | LightGBM | Machine Learning | 0.78 | 0.76 | 0.96 | 0.64 |
-| 10 | FT Transformer | Deep Learning / Transformer | 0.76 | 0.76 | 0.94 | 0.61 |
-""",
-}
-
-PERFORMANCE_GUIDE_FULL = {
-    "printability": """
-# Printability - Merged Performance Summary
-
-This file contains the aggregated and benchmarked results of traditional Machine Learning (ML) and Deep Learning / Transformer architectures for predicting **Printability**, sorted hierarchically by **F1 Score** and **Accuracy**.
-
-| Rank | Model | Framework | Accuracy | Precision | Recall | F1 | AUC | MCC | Kappa |
-|---:|---|---|---:|---:|---:|---:|---:|---:|---:|
-| 1 | HistGradientBoosting | Machine Learning | 0.80 | 0.80 | 0.80 | 0.80 | 0.94 | 0.69 | 0.69 |
-| 2 | TabPFN 2.6 | Deep Learning / Transformer | 0.80 | 0.81 | 0.80 | 0.80 | 0.94 | 0.69 | 0.68 |
-| 3 | Bagging | Machine Learning | 0.79 | 0.79 | 0.79 | 0.79 | 0.93 | 0.67 | 0.67 |
-| 4 | GradientBoosting | Machine Learning | 0.79 | 0.79 | 0.79 | 0.79 | 0.93 | 0.66 | 0.66 |
-| 5 | TabICL v2 | Deep Learning / Transformer | 0.79 | 0.81 | 0.79 | 0.79 | 0.94 | 0.68 | 0.67 |
-| 6 | XGBoost | Machine Learning | 0.78 | 0.79 | 0.78 | 0.78 | 0.93 | 0.66 | 0.65 |
-| 7 | LightGBM | Machine Learning | 0.77 | 0.78 | 0.77 | 0.77 | 0.93 | 0.64 | 0.64 |
-| 8 | KNeighbors | Machine Learning | 0.77 | 0.78 | 0.77 | 0.77 | 0.90 | 0.64 | 0.64 |
-| 9 | LabelPropagation | Machine Learning | 0.77 | 0.77 | 0.77 | 0.77 | 0.82 | 0.64 | 0.63 |
-| 10 | ExtraTrees | Machine Learning | 0.77 | 0.77 | 0.77 | 0.77 | 0.92 | 0.63 | 0.63 |
-| 11 | MLP | Machine Learning | 0.76 | 0.76 | 0.76 | 0.76 | 0.90 | 0.63 | 0.63 |
-| 12 | LabelSpreading | Machine Learning | 0.76 | 0.76 | 0.76 | 0.76 | 0.90 | 0.62 | 0.62 |
-| 13 | TabNet Lite | Deep Learning / Transformer | 0.74 | 0.74 | 0.74 | 0.74 | 0.91 | 0.59 | 0.59 |
-| 14 | MLP (DL) | Deep Learning / Transformer | 0.73 | 0.73 | 0.73 | 0.73 | 0.91 | 0.58 | 0.58 |
-| 15 | ResNet | Deep Learning / Transformer | 0.73 | 0.73 | 0.73 | 0.73 | 0.91 | 0.57 | 0.57 |
-| 16 | 1D CNN | Deep Learning / Transformer | 0.71 | 0.73 | 0.71 | 0.72 | 0.89 | 0.56 | 0.56 |
-| 17 | DecisionTree | Machine Learning | 0.71 | 0.72 | 0.71 | 0.71 | 0.89 | 0.54 | 0.54 |
-| 18 | NODE Lite | Deep Learning / Transformer | 0.69 | 0.70 | 0.69 | 0.70 | 0.90 | 0.52 | 0.52 |
-| 19 | LinearSVC | Machine Learning | 0.69 | 0.68 | 0.69 | 0.68 | 0.85 | 0.49 | 0.49 |
-| 20 | LDA | Machine Learning | 0.68 | 0.68 | 0.68 | 0.68 | 0.84 | 0.49 | 0.48 |
-| 21 | PassiveAggressive | Machine Learning | 0.69 | 0.68 | 0.69 | 0.67 | 0.85 | 0.48 | 0.48 |
-| 22 | CalibratedClassifierCV | Machine Learning | 0.68 | 0.68 | 0.68 | 0.67 | 0.85 | 0.48 | 0.48 |
-| 23 | LogisticRegression | Machine Learning | 0.68 | 0.67 | 0.68 | 0.67 | 0.85 | 0.47 | 0.47 |
-| 24 | RidgeClassifier | Machine Learning | 0.68 | 0.67 | 0.68 | 0.66 | 0.84 | 0.46 | 0.46 |
-| 25 | FT Transformer | Deep Learning / Transformer | 0.66 | 0.65 | 0.66 | 0.65 | 0.87 | 0.45 | 0.45 |
-| 26 | Perceptron | Machine Learning | 0.67 | 0.68 | 0.67 | 0.64 | 0.84 | 0.45 | 0.42 |
-| 27 | MultinomialNB | Machine Learning | 0.64 | 0.69 | 0.64 | 0.64 | 0.83 | 0.45 | 0.43 |
-| 28 | RadiusNeighbors | Machine Learning | 0.66 | 0.70 | 0.66 | 0.63 | 0.87 | 0.43 | 0.39 |
-| 29 | ComplementNB | Machine Learning | 0.63 | 0.67 | 0.63 | 0.63 | 0.82 | 0.44 | 0.43 |
-| 30 | NuSVC | Machine Learning | 0.60 | 0.69 | 0.60 | 0.61 | 0.84 | 0.45 | 0.42 |
-| 31 | AdaBoost | Machine Learning | 0.62 | 0.59 | 0.62 | 0.58 | 0.80 | 0.38 | 0.37 |
-| 32 | SGD | Machine Learning | 0.63 | 0.59 | 0.63 | 0.57 | 0.82 | 0.36 | 0.35 |
-| 33 | BernoulliNB | Machine Learning | 0.59 | 0.62 | 0.59 | 0.55 | 0.77 | 0.32 | 0.31 |
-| 34 | GaussianNB | Machine Learning | 0.33 | 0.70 | 0.33 | 0.36 | 0.74 | 0.25 | 0.19 |
-| 35 | QDA | Machine Learning | 0.52 | 0.27 | 0.52 | 0.35 | 0.80 | 0.00 | 0.00 |
-| 36 | RandomForest | Machine Learning | 0.52 | 0.27 | 0.52 | 0.35 | 0.78 | 0.00 | 0.00 |
-| 37 | ExtraTreeClassifier | Machine Learning | 0.52 | 0.27 | 0.52 | 0.35 | 0.50 | 0.00 | 0.00 |
-| 38 | DummyClassifier | Machine Learning | 0.52 | 0.27 | 0.52 | 0.35 | 0.50 | 0.00 | 0.00 |
-""",
-    "cell_response": """
-# Cell Response - Merged Performance Summary
-
-This file contains the aggregated and benchmarked results of traditional Machine Learning (ML) and Deep Learning / Transformer architectures for predicting **Cell Response**, sorted hierarchically by **F1 Score** and **Accuracy**.
-
-| Rank | Model | Framework | Accuracy | Precision | Recall | F1 | AUC | MCC | Kappa |
-|---:|---|---|---:|---:|---:|---:|---:|---:|---:|
-| 1 | HistGradientBoosting | Machine Learning | 0.81 | 0.81 | 0.81 | 0.81 | 0.96 | 0.67 | 0.67 |
-| 2 | TabICL v2 | Deep Learning / Transformer | 0.79 | 0.79 | 0.79 | 0.79 | 0.97 | 0.65 | 0.65 |
-| 3 | TabPFN 2.6 | Deep Learning / Transformer | 0.78 | 0.78 | 0.78 | 0.78 | 0.96 | 0.63 | 0.63 |
-| 4 | XGBoost | Machine Learning | 0.79 | 0.76 | 0.79 | 0.77 | 0.96 | 0.65 | 0.64 |
-| 5 | Bagging | Machine Learning | 0.77 | 0.78 | 0.77 | 0.77 | 0.95 | 0.62 | 0.62 |
-| 6 | KNeighbors | Machine Learning | 0.77 | 0.77 | 0.77 | 0.77 | 0.93 | 0.61 | 0.61 |
-| 7 | LabelPropagation | Machine Learning | 0.77 | 0.77 | 0.77 | 0.77 | 0.93 | 0.61 | 0.61 |
-| 8 | LabelSpreading | Machine Learning | 0.78 | 0.76 | 0.78 | 0.76 | 0.93 | 0.61 | 0.61 |
-| 9 | LightGBM | Machine Learning | 0.78 | 0.79 | 0.78 | 0.76 | 0.96 | 0.64 | 0.63 |
-| 10 | FT Transformer | Deep Learning / Transformer | 0.76 | 0.79 | 0.76 | 0.76 | 0.94 | 0.61 | 0.60 |
-| 11 | CalibratedClassifierCV | Machine Learning | 0.76 | 0.76 | 0.76 | 0.75 | 0.94 | 0.59 | 0.59 |
-| 12 | ResNet | Deep Learning / Transformer | 0.76 | 0.76 | 0.76 | 0.75 | 0.94 | 0.59 | 0.59 |
-| 13 | NODE Lite | Deep Learning / Transformer | 0.76 | 0.75 | 0.76 | 0.75 | 0.95 | 0.59 | 0.59 |
-| 14 | 1D CNN | Deep Learning / Transformer | 0.75 | 0.76 | 0.75 | 0.74 | 0.95 | 0.59 | 0.59 |
-| 15 | MLP (DL) | Deep Learning / Transformer | 0.75 | 0.75 | 0.75 | 0.74 | 0.95 | 0.57 | 0.57 |
-| 16 | LDA | Machine Learning | 0.75 | 0.72 | 0.75 | 0.73 | 0.94 | 0.55 | 0.55 |
-| 17 | NuSVC | Machine Learning | 0.74 | 0.73 | 0.74 | 0.73 | 0.95 | 0.55 | 0.55 |
-| 18 | MultinomialNB | Machine Learning | 0.74 | 0.72 | 0.74 | 0.73 | 0.95 | 0.55 | 0.55 |
-| 19 | LogisticRegression | Machine Learning | 0.74 | 0.70 | 0.74 | 0.71 | 0.95 | 0.54 | 0.54 |
-| 20 | LinearSVC | Machine Learning | 0.74 | 0.71 | 0.74 | 0.70 | 0.94 | 0.53 | 0.52 |
-| 21 | Perceptron | Machine Learning | 0.74 | 0.68 | 0.74 | 0.70 | 0.94 | 0.53 | 0.52 |
-| 22 | QDA | Machine Learning | 0.74 | 0.68 | 0.74 | 0.70 | 0.94 | 0.57 | 0.55 |
-| 23 | RidgeClassifier | Machine Learning | 0.73 | 0.70 | 0.73 | 0.70 | 0.94 | 0.52 | 0.51 |
-| 24 | ExtraTrees | Machine Learning | 0.74 | 0.71 | 0.74 | 0.68 | 0.95 | 0.55 | 0.52 |
-| 25 | ExtraTreeClassifier | Machine Learning | 0.73 | 0.67 | 0.73 | 0.68 | 0.91 | 0.54 | 0.52 |
-| 26 | AdaBoost | Machine Learning | 0.74 | 0.65 | 0.74 | 0.67 | 0.91 | 0.59 | 0.55 |
-| 27 | DecisionTree | Machine Learning | 0.74 | 0.65 | 0.74 | 0.67 | 0.92 | 0.59 | 0.55 |
-| 28 | RandomForest | Machine Learning | 0.74 | 0.65 | 0.74 | 0.67 | 0.92 | 0.59 | 0.55 |
-| 29 | GradientBoosting | Machine Learning | 0.74 | 0.65 | 0.74 | 0.67 | 0.92 | 0.59 | 0.55 |
-| 30 | SGD | Machine Learning | 0.71 | 0.65 | 0.71 | 0.66 | 0.92 | 0.49 | 0.47 |
-| 31 | MLP | Machine Learning | 0.73 | 0.61 | 0.73 | 0.65 | 0.93 | 0.55 | 0.52 |
-| 32 | PassiveAggressive | Machine Learning | 0.73 | 0.65 | 0.73 | 0.65 | 0.92 | 0.52 | 0.49 |
-| 33 | ComplementNB | Machine Learning | 0.72 | 0.65 | 0.72 | 0.65 | 0.93 | 0.49 | 0.47 |
-| 34 | TabNet Lite | Deep Learning / Transformer | 0.70 | 0.67 | 0.70 | 0.65 | 0.93 | 0.48 | 0.46 |
-| 35 | BernoulliNB | Machine Learning | 0.71 | 0.60 | 0.71 | 0.63 | 0.93 | 0.48 | 0.45 |
-| 36 | RadiusNeighbors | Machine Learning | 0.60 | 0.36 | 0.60 | 0.45 | 0.50 | 0.00 | 0.00 |
-| 37 | DummyClassifier | Machine Learning | 0.60 | 0.36 | 0.60 | 0.45 | 0.50 | 0.00 | 0.00 |
-| 38 | GaussianNB | Machine Learning | 0.28 | 0.69 | 0.28 | 0.34 | 0.84 | 0.18 | 0.14 |
-""",
-}
-
-GEMINI_MODELS = [
-    "gemini-3.5-flash",
-    "gemini-3.1-flash-lite",
-    "gemini-3.1-pro-preview",
-    "gemini-3.1-flash-lite-preview",
-    "gemini-3-flash-preview",
-    "gemini-2.5-pro",
-    "gemini-2.5-flash",
-    "gemini-2.5-flash-lite",
-]
-
-def scaffold_quality_combined(printability, cell_response, 
-                              weight_printability=0.3, weight_cell_response=0.7):
-    """
-    Calculates the Weighted Scaffold Synthesis Quality (WSSQ).
-    """
-    if printability == 0:
-        return 0.0
-    
-    # Normalization
-    norm_p = printability / 3.0
-    
-    # If cell_response is 1 (minimum), avoid division by zero in harmonic mean
-    if cell_response <= 1:
-        return 100 * norm_p
-    
-    norm_c = (cell_response - 1) / 4.0
-    
-    # Weighted Harmonic Mean
-    hm = (weight_printability + weight_cell_response) / (
-        (weight_printability / norm_p) + 
-        (weight_cell_response / norm_c)
-    )
-    
-    # Weighted Multiplicative Component
-    mc = (norm_p**weight_printability) * (norm_c**weight_cell_response)
-    
-    return 100 * ((hm + mc) / 2.0)
-
-PRINT_PARAM_NAMES = [
-    "Physical Crosslinking Duration (s)",
-    "Photo Crosslinking Duration (s)",
-    "Extrusion Pressure (kPa)",
-    "Nozzle Movement Speed (mm/s)",
-    "Nozzle Diameter (µm)",
-    "Syringe Temperature (°C)",
-    "Substrate Temperature (°C)",
-]
-
-@st.cache_resource
-def load_prediction_preprocessors():
-    return {
-        "preprocessor": joblib.load(PREPROCESSOR_DIR / "preprocessor.pkl"),
-        "feature_cols": joblib.load(PREPROCESSOR_DIR / "feature_cols.pkl"),
-        "printability_encoder": joblib.load(PREPROCESSOR_DIR / MODEL_TASKS["printability"]["label_encoder"]),
-        "cell_response_encoder": joblib.load(PREPROCESSOR_DIR / MODEL_TASKS["cell_response"]["label_encoder"]),
-    }
-
-def model_display_name(path, prefix):
-    name = path.stem.replace("_model", "")
-    if name.startswith(prefix):
-        name = name[len(prefix):]
-    return f"{name} ({path.suffix.lstrip('.')})"
-
-def model_rank_key(display_name, task_key):
-    base_name = display_name.rsplit(" (", 1)[0]
-    rank_aliases = {
-        "1D_CNN": "1D CNN",
-        "FT_Transformer": "FT Transformer",
-        "NODE_Lite": "NODE Lite",
-        "TabNet_Lite": "TabNet Lite",
-        "TabPFN_2.6": "TabPFN 2.6",
-        "TabICL_v2": "TabICL v2",
-        "ExtraTree": "ExtraTreeClassifier",
-    }
-    rank_name = rank_aliases.get(base_name, base_name)
-    if base_name == "MLP" and display_name.endswith("(pth)"):
-        rank_name = "MLP DL"
-    ranking = MODEL_RANKINGS[task_key]
-    rank = ranking.index(rank_name) if rank_name in ranking else len(ranking)
-    return rank, base_name.lower(), display_name
-
-def discover_model_options(task_key):
-    task = MODEL_TASKS[task_key]
-    files = []
-    for suffix in ("*.pkl", "*.joblib", "*.pth"):
-        files.extend(task["folder"].glob(suffix))
-    options = {model_display_name(path, task["prefix"]): str(path) for path in files}
-    return dict(sorted(options.items(), key=lambda item: model_rank_key(item[0], task_key)))
-
-def parse_architecture(path, task_key):
-    stem = Path(path).stem.replace("_model", "")
-    prefix = MODEL_TASKS[task_key]["prefix"]
-    return stem[len(prefix):] if stem.startswith(prefix) else stem
-
-def build_torch_model(architecture, input_dim, out_dim, cfg):
-    import torch
-    import torch.nn as nn
-
-    activation_funcs = {
-        "relu": nn.ReLU, "tanh": nn.Tanh, "GELU": nn.GELU,
-        "SELU": nn.SELU, "ELU": nn.ELU, "SiLU": nn.SiLU,
-    }
-    act = activation_funcs[cfg["activation_func"]]
-
-    class ResidualBlock(nn.Module):
-        def __init__(self):
-            super().__init__()
-            self.linear = nn.Linear(cfg["hidden_dim"], cfg["hidden_dim"])
-            self.bn = nn.BatchNorm1d(cfg["hidden_dim"])
-            self.act = act()
-            self.dropout = nn.Dropout(cfg["dropout"])
-        def forward(self, x):
-            return x + self.dropout(self.act(self.bn(self.linear(x))))
-
-    class TissueResNet(nn.Module):
-        def __init__(self):
-            super().__init__()
-            self.input_layer = nn.Sequential(
-                nn.Linear(input_dim, cfg["hidden_dim"]),
-                nn.BatchNorm1d(cfg["hidden_dim"]),
-                act(),
-            )
-            self.blocks = nn.ModuleList([ResidualBlock() for _ in range(cfg["n_layers"])])
-            self.output_layer = nn.Linear(cfg["hidden_dim"], out_dim)
-        def forward(self, x):
-            x = self.input_layer(x)
-            for block in self.blocks:
-                x = block(x)
-            return self.output_layer(x)
-
-    class StandardMLP(nn.Module):
-        def __init__(self):
-            super().__init__()
-            layers = []
-            in_dim = input_dim
-            for _ in range(cfg["n_layers"]):
-                layers.extend([
-                    nn.Linear(in_dim, cfg["hidden_dim"]),
-                    nn.BatchNorm1d(cfg["hidden_dim"]),
-                    act(),
-                    nn.Dropout(cfg["dropout"]),
-                ])
-                in_dim = cfg["hidden_dim"]
-            layers.append(nn.Linear(cfg["hidden_dim"], out_dim))
-            self.network = nn.Sequential(*layers)
-        def forward(self, x):
-            return self.network(x)
-
-    class Tabular1DCNN(nn.Module):
-        def __init__(self):
-            super().__init__()
-            layers = []
-            in_channels = 1
-            for _ in range(cfg["n_layers"]):
-                layers.extend([
-                    nn.Conv1d(in_channels, cfg["hidden_dim"], kernel_size=3, padding=1),
-                    nn.BatchNorm1d(cfg["hidden_dim"]),
-                    act(),
-                    nn.Dropout(cfg["dropout"]),
-                ])
-                in_channels = cfg["hidden_dim"]
-            self.conv_net = nn.Sequential(*layers)
-            self.pool = nn.AdaptiveAvgPool1d(1)
-            self.fc = nn.Linear(cfg["hidden_dim"], out_dim)
-        def forward(self, x):
-            x = self.conv_net(x.unsqueeze(1))
-            return self.fc(self.pool(x).squeeze(2))
-
-    class FTTransformer(nn.Module):
-        def __init__(self):
-            super().__init__()
-            self.d_token = max(4, (cfg["hidden_dim"] // 4) * 4)
-            self.feature_embeddings = nn.ModuleList([nn.Linear(1, self.d_token) for _ in range(input_dim)])
-            self.cls_token = nn.Parameter(torch.randn(1, 1, self.d_token))
-            encoder_layer = nn.TransformerEncoderLayer(
-                d_model=self.d_token, nhead=4, dropout=cfg["dropout"], batch_first=True
-            )
-            self.transformer = nn.TransformerEncoder(encoder_layer, num_layers=cfg["n_layers"])
-            self.fc = nn.Linear(self.d_token, out_dim)
-        def forward(self, x):
-            batch_size = x.size(0)
-            tokens = [self.feature_embeddings[i](x[:, i:i+1]).unsqueeze(1) for i in range(x.size(1))]
-            x_emb = torch.cat([self.cls_token.expand(batch_size, -1, -1)] + tokens, dim=1)
-            return self.fc(self.transformer(x_emb)[:, 0, :])
-
-    class TabNetLite(nn.Module):
-        def __init__(self):
-            super().__init__()
-            self.n_steps = max(1, cfg["n_layers"])
-            self.initial_bn = nn.BatchNorm1d(input_dim)
-            self.transformers = nn.ModuleList([
-                nn.Sequential(
-                    nn.Linear(input_dim, cfg["hidden_dim"]),
-                    nn.BatchNorm1d(cfg["hidden_dim"]),
-                    act(),
-                    nn.Dropout(cfg["dropout"]),
-                )
-                for _ in range(self.n_steps)
-            ])
-            self.attentions = nn.ModuleList([
-                nn.Sequential(
-                    nn.Linear(cfg["hidden_dim"], input_dim),
-                    nn.BatchNorm1d(input_dim),
-                    nn.Softmax(dim=-1),
-                )
-                for _ in range(self.n_steps)
-            ])
-            self.fc_out = nn.Linear(cfg["hidden_dim"], out_dim)
-        def forward(self, x):
-            x = self.initial_bn(x)
-            out_agg, prior = 0, torch.ones_like(x)
-            feat_rep = self.transformers[0](x)
-            for step in range(self.n_steps):
-                mask = self.attentions[step](feat_rep) * prior
-                prior = prior * (1.0 - mask)
-                feat_rep = self.transformers[step](x * mask)
-                out_agg += feat_rep
-            return self.fc_out(out_agg)
-
-    class NeuralDecisionForest(nn.Module):
-        def __init__(self):
-            super().__init__()
-            self.n_trees = max(1, cfg["hidden_dim"] // 16)
-            self.depth = max(2, cfg["n_layers"] + 1)
-            self.n_leaves = 2 ** self.depth
-            self.trees = nn.ModuleList([
-                nn.Sequential(
-                    nn.Linear(input_dim, self.n_leaves),
-                    nn.Dropout(cfg["dropout"]),
-                    nn.Softmax(dim=-1),
-                )
-                for _ in range(self.n_trees)
-            ])
-            self.leaf_weights = nn.Parameter(torch.randn(self.n_trees, self.n_leaves, out_dim))
-        def forward(self, x):
-            out = 0
-            for i, tree in enumerate(self.trees):
-                out += torch.matmul(tree(x), self.leaf_weights[i])
-            return out / self.n_trees
-
-    builders = {
-        "ResNet": TissueResNet,
-        "MLP": StandardMLP,
-        "1D_CNN": Tabular1DCNN,
-        "FT_Transformer": FTTransformer,
-        "TabNet_Lite": TabNetLite,
-        "NODE_Lite": NeuralDecisionForest,
-    }
-    return builders[architecture]()
-
-@st.cache_resource
-def load_prediction_model(path, task_key, input_dim, out_dim):
-    path = Path(path)
-    if path.suffix in {".pkl", ".joblib"}:
-        return {"kind": "sklearn", "model": joblib.load(path)}
-
-    if path.suffix == ".pth":
-        import torch
-        architecture = parse_architecture(path, task_key)
-        cfg = DL_MODEL_CONFIGS.get((task_key, architecture))
-        if cfg is None:
-            raise ValueError(f"No architecture configuration found for {path.name}.")
-        model = build_torch_model(architecture, input_dim, out_dim, cfg)
-        state_dict = torch.load(path, map_location="cpu")
-        model.load_state_dict(state_dict)
-        model.eval()
-        return {"kind": "torch", "model": model}
-
-    raise ValueError(f"Unsupported model file: {path.name}")
-
-def softmax(values):
-    values = np.asarray(values, dtype=float)
-    values = values - np.max(values)
-    exp_values = np.exp(values)
-    return exp_values / exp_values.sum()
-
-def expected_class_value(model_bundle, x_raw, preprocessor, label_encoder):
-    x_model = preprocessor.transform(x_raw).astype(np.float32)
-
-    if model_bundle["kind"] == "torch":
-        import torch
-        with torch.no_grad():
-            logits = model_bundle["model"](torch.tensor(x_model, dtype=torch.float32)).numpy()[0]
-        probs = softmax(logits)
-        labels = label_encoder.classes_.astype(float)
-        return float(np.dot(probs, labels))
-
-    model = model_bundle["model"]
-    if hasattr(model, "predict_proba"):
-        probs = model.predict_proba(x_model)[0]
-        classes = np.asarray(model.classes_, dtype=int)
-        labels = label_encoder.inverse_transform(classes).astype(float)
-        return float(np.dot(probs, labels))
-
-    if hasattr(model, "decision_function"):
-        scores = np.asarray(model.decision_function(x_model)[0])
-        if scores.ndim == 0:
-            p_high = 1.0 / (1.0 + np.exp(-scores))
-            probs = np.array([1.0 - p_high, p_high])
-        else:
-            probs = softmax(scores)
-        classes = np.asarray(model.classes_, dtype=int)
-        labels = label_encoder.inverse_transform(classes).astype(float)
-        return float(np.dot(probs, labels))
-
-    pred = np.asarray(model.predict(x_model), dtype=int)
-    return float(label_encoder.inverse_transform(pred)[0])
-
-prediction_assets = load_prediction_preprocessors()
-preprocessor = prediction_assets["preprocessor"]
-feature_cols = prediction_assets["feature_cols"]
-label_encoder_print = prediction_assets["printability_encoder"]
-label_encoder_cell = prediction_assets["cell_response_encoder"]
-
-sample_for_shape = {col: 0.0 for col in feature_cols}
-sample_for_shape["Cell Line"] = CELL_LINE_OPTIONS[0]
-preprocessed_input_dim = preprocessor.transform(pd.DataFrame([sample_for_shape])[feature_cols]).shape[1]
-
-@st.dialog("Optimization Trials")
-def show_trial_guidance():
-    st.markdown(
-        """
-        The trial count controls how many candidate scaffold settings Optuna tests before choosing the best WSSQ.
-
-        **Recommended range:** 100-1000 trials.
-
-        **Default:** 300 trials, which is a balanced choice for normal use.
-
-        **Runtime impact:** running time grows roughly in proportion to the number of trials. Use 50-100 for a quick test, 300 for a balanced run, and 500-1000 when you want a more thorough search and can wait longer.
-        """
-    )
-
-@st.dialog("Weighted Synergistic Scaffold Quality (WSSQ)", width="large")
-def show_wssq_guidance():
-    st.markdown(
-        """
-        WSSQ is the optimization score used in MLATE to combine **printability** and **cell response** into one scaffold-quality objective.
-
-        WSSQ was introduced to handle two practical needs: acellular 3D-printed scaffolds, where cell response is not applicable in the same way, and bioprinted scaffolds, where biological response is central to scaffold quality. 
-
-        The app normalizes printability and cell response, then combines them using two components:
-
-        - **Weighted harmonic mean:** rewards balanced high values and penalizes weak performance in either target.
-        - **Weighted multiplicative component:** captures synergy between printability and cell response.
-
-        The final WSSQ is scaled from 0 to 100%. If printability is 0, WSSQ is 0. If cell response is at the minimum biological-response level, the score falls back to normalized printability. Otherwise, both targets contribute according to the sidebar weights.
-
-        Practically, a scaffold with excellent cell response but poor printability can score lower than a scaffold with slightly lower cell response but better balance, because WSSQ is designed to favor experimentally useful, well-balanced scaffold candidates.
-        """
-    )
-
-@st.dialog("Model Selection Guide", width="large")
-def show_model_selection_guidance():
-    st.markdown(
-        """
-        Models are ranked by F1 score and accuracy. The highest-ranked available model appears first in each dropdown.
-
-        Use the top-ranked models when you want the strongest benchmarked predictive performance. If a selected model fails to load because of local package-version incompatibility, choose the next ranked model in the same task until the environment is aligned with the model artifacts.
-        """
-    )
-    tab_print, tab_cell = st.tabs(["Printability", "Cell Response"])
-    with tab_print:
-        st.markdown(PERFORMANCE_GUIDE["printability"])
-        with st.expander("Show more"):
-            st.markdown(PERFORMANCE_GUIDE_FULL["printability"])
-    with tab_cell:
-        st.markdown(PERFORMANCE_GUIDE["cell_response"])
-        with st.expander("Show more"):
-            st.markdown(PERFORMANCE_GUIDE_FULL["cell_response"])
-
-@st.dialog("How to Get a Gemini API Key")
-def show_gemini_api_key_guidance():
-    st.markdown(
-        """
-        To generate a fabrication procedure, you need a Gemini API key from Google. Creating a key only takes a minute, and Google provides a free tier.
-
-        1. Open [Google AI Studio API Keys](https://aistudio.google.com/app/apikey) and sign in with your Google/Gmail account if prompted.
-        2. If this is your first visit, accept the terms of service and continue.
-        3. Click **Get API key** or **Create API key**.
-        4. Choose an existing Google Cloud project, or select **Create API key in new project**.
-        5. Copy the generated key, return to this app, and paste it into the **Gemini API Key** box.
-
-        **Important:** Treat your API key like a password. Do not share it publicly or paste it into files that will be uploaded online. This app uses your key only for the current protocol generation request and does not save it.
-        """
-    )
-
-if 'bio_rows' not in st.session_state:
-    st.session_state.bio_rows = [{
-        'mat': BIOMATERIAL_OPTIONS[0],
-        'min': 0.0, 'max': 10.0, 'step': 0.1
-    }]
-
-if 'density_range' not in st.session_state:
-    st.session_state.density_range = {'min': 0.0, 'max': 10.0, 'step': 0.1}
-
-if 'pp_ranges' not in st.session_state:
-    st.session_state.pp_ranges = {
-        "Physical Crosslinking Duration (s)": {'min': 0.0, 'max': 300.0, 'step': 5.0},
-        "Photo Crosslinking Duration (s)":    {'min': 0.0, 'max': 180.0,  'step': 5.0},
-        "Extrusion Pressure (kPa)":           {'min': 5.0,  'max': 200.0,  'step': 5.0},
-        "Nozzle Movement Speed (mm/s)":       {'min': 1.0,  'max': 20.0,   'step': 0.5},
-        "Nozzle Diameter (µm)":               {'min': 100.0,'max': 1000.0, 'step': 50.0},
-        "Syringe Temperature (°C)":           {'min': 4.0,  'max': 40.0,   'step': 1.0},
-        "Substrate Temperature (°C)":         {'min': 4.0,  'max': 37.0,   'step': 1.0},
-    }
-
-# --- Sidebar UI for Weights ---
-st.sidebar.header("Optimization Weights")
-# User only controls Cell Response (0 to 100)
-w_cell_pct = st.sidebar.slider("Cell Response Weight (%)", min_value=0, max_value=100, value=70, step=5)
-# Printability is dynamically calculated and cannot be changed manually
-w_print_pct = 100 - w_cell_pct
-st.sidebar.number_input("Printability Weight (%)", value=w_print_pct, disabled=True, help="Auto-calculated to ensure sum is 100%")
-if st.sidebar.button("What is WSSQ?", use_container_width=True):
-    show_wssq_guidance()
-
-# Convert back to 0.0 - 1.0 for the mathematical formula
-w_cell = w_cell_pct / 100.0
-w_print = w_print_pct / 100.0
-
-print_model_options = discover_model_options("printability")
-cell_model_options = discover_model_options("cell_response")
-
-if not print_model_options or not cell_model_options:
-    st.error("No selectable prediction models were found in the models folder.")
-    st.stop()
-
-selected_print_model_name = st.sidebar.selectbox(
-    "Printability Model",
-    list(print_model_options.keys()),
-    key="printability_model_select",
-)
-selected_cell_model_name = st.sidebar.selectbox(
-    "Cell Response Model",
-    list(cell_model_options.keys()),
-    key="cell_response_model_select",
-)
-if st.sidebar.button("Model Selection Guide", use_container_width=True):
-    show_model_selection_guidance()
-
-n_trials = st.sidebar.number_input(
-    "Optimization Trials",
-    min_value=10,
-    max_value=10000,
-    value=300,
-    step=50,
-    help="Number of Optuna trials used when you click Optimize WSSQ.",
-)
-if st.sidebar.button("Trial Count Help", use_container_width=True):
-    show_trial_guidance()
-
-gemini_key = st.sidebar.text_input(
-    "Gemini API Key",
-    value=os.getenv("GEMINI_API_KEY", ""),
-    type="password",
-    help="Used only when generating the LLM-based fabrication procedure.",
-)
-if st.sidebar.button("How to Get API Key", use_container_width=True):
-    show_gemini_api_key_guidance()
-
-gemini_model = st.sidebar.selectbox(
-    "Gemini Model",
-    GEMINI_MODELS,
-    index=0,
-    key="gemini_model_select",
-)
-
-st.title("MLATE: Machine Learning Applications in Tissue Engineering")
-st.markdown(
-    "<p style='font-size:1.2em; color:grey;'>"
-    "A Data-driven Cross-tissue Machine Learning Framework for Inverse design of 3D (Bio)printing Scaffolds "
-    "For more details, please refer to and cite our paper: "
-    "<a href='https://doi.org/xxx' target='_blank'>https://doi.org/xxx</a>"
-    "</p>",
-    unsafe_allow_html=True
-)
-
-st.subheader("Biomaterials (enter range for each)")
-if st.button("➕ Add Biomaterial"):
-    used = {r['mat'] for r in st.session_state.bio_rows}
-    available = [m for m in BIOMATERIAL_OPTIONS if m not in used]
-    if available:
-        st.session_state.bio_rows.append({
-            'mat': available[0], 'min': 0.0, 'max': 10.0, 'step': 0.1
-        })
-    st.rerun()
-
-for i, row in enumerate(st.session_state.bio_rows):
-    used_except_current = {
-        r['mat'] for idx, r in enumerate(st.session_state.bio_rows) if idx != i
-    }
-    options = [m for m in BIOMATERIAL_OPTIONS if m not in used_except_current]
-
-    c1, c2, c3, c4, c5 = st.columns([2, 1, 1, 1, 0.3])
-    mat = c1.selectbox(
-        "Biomaterial", options,
-        index=options.index(row['mat']) if row['mat'] in options else 0,
-        key=f"bio_mat_{i}",
-        label_visibility="collapsed",
-    )
-    st.session_state.bio_rows[i]['mat'] = mat
-
-    mn = c2.number_input(
-        "Min", min_value=0.0, max_value=row['max'],
-        value=row['min'], step=row['step'], key=f"bio_min_{i}"
-    )
-    mx = c3.number_input(
-        "Max", min_value=row['step'], max_value=100.0,
-        value=max(row['max'], row['step']), step=row['step'],
-        key=f"bio_max_{i}"
-    )
-    st.session_state.bio_rows[i].update(min=mn, max=mx)
-
-    st.session_state.bio_rows[i]['step'] = c4.number_input(
-        "Step", min_value=0.0,
-        max_value=(mx - mn) if mx > mn else 0.1,
-        value=row['step'], step=0.1, key=f"bio_step_{i}"
-    )
-
-    if c5.button("❌", key=f"rem_{i}"):
-        st.session_state.bio_rows.pop(i)
-        st.rerun()
-
-st.markdown("---")
-
-st.subheader("Cell Line & Density (10^6 cells/ml)")
-col1, col2, col3, col4 = st.columns([2,1,1,1])
-
-cell_line = col1.selectbox("Cell Line", CELL_LINE_OPTIONS, key="cell_line_select")
-
-if cell_line == "NoCellCultured":
-    st.info("🧪 **Acellular 3D printing mode** – No cells will be included. Cell density is forced to 0.")
-    st.session_state.density_range.update({'min': 0.0, 'max': 0.0, 'step': 0.0})
-    
-    col2.number_input("Min Density", value=0.0, disabled=True, key="cd_min")
-    col3.number_input("Max Density", value=0.0, disabled=True, key="cd_max")
-    col4.number_input("Step", value=0.0, disabled=True, key="cd_step")
-else:
-    if st.session_state.density_range.get('max', 0) <= 0.1:
-        st.session_state.density_range.update({'min': 1.0, 'max': 20.0, 'step': 0.5})
-    
-    dr = st.session_state.density_range
-    dmin = col2.number_input(
-        "Min Density", 
-        min_value=0.0, 
-        max_value=dr['max'],
-        value=dr['min'], 
-        step=dr['step'], 
-        key="cd_min"
-    )
-    dmax = col3.number_input(
-        "Max Density", 
-        min_value=dr['step'], 
-        max_value=1000.0,
-        value=max(dr['max'], dr['step']), 
-        step=dr['step'], 
-        key="cd_max"
-    )
-    dstep = col4.number_input(
-        "Step", 
-        min_value=0.0,
-        max_value=(dmax - dmin) if dmax > dmin else 0.1,
-        value=dr['step'], 
-        step=0.1, 
-        key="cd_step"
-    )
-    st.session_state.density_range.update({'min': dmin, 'max': dmax, 'step': dstep})
-    
-st.markdown("---")
-
-st.subheader("Crosslinking Settings")
-
-col_cross1, col_cross2 = st.columns(2)
-
-disable_physical = col_cross1.checkbox(
-    "Disable Physical Crosslinking", 
-    value=False, 
-    help="Check if you do not want physical/ionic crosslinking (e.g. CaCl₂ bath, temperature-induced)"
-)
-
-disable_photo = col_cross2.checkbox(
-    "Disable Photo Crosslinking", 
-    value=False, 
-    help="Check if you do not want UV/visible light crosslinking"
-)
-
-st.subheader("Printing Parameters (enter range)")
-
-for name in PRINT_PARAM_NAMES:
-    if name == "Physical Crosslinking Duration (s)" and disable_physical:
-        st.session_state.pp_ranges[name].update({'min': 0.0, 'max': 0.0, 'step': 0.0})
-        c1, c2, c3, c4 = st.columns([2,1,1,1])
-        c1.write(name + " (DISABLED)")
-        c2.number_input("Min", value=0.0, disabled=True, key=f"pp_min_{name}")
-        c3.number_input("Max", value=0.0, disabled=True, key=f"pp_max_{name}")
-        c4.number_input("Step", value=0.0, disabled=True, key=f"pp_step_{name}")
-        continue
-    
-    elif name == "Photo Crosslinking Duration (s)" and disable_photo:
-        st.session_state.pp_ranges[name].update({'min': 0.0, 'max': 0.0, 'step': 0.0})
-        c1, c2, c3, c4 = st.columns([2,1,1,1])
-        c1.write(name + " (DISABLED)")
-        c2.number_input("Min", value=0.0, disabled=True, key=f"pp_min_{name}")
-        c3.number_input("Max", value=0.0, disabled=True, key=f"pp_max_{name}")
-        c4.number_input("Step", value=0.0, disabled=True, key=f"pp_step_{name}")
-        continue
-
-    pmin = st.session_state.pp_ranges[name]['min']
-    pmax = st.session_state.pp_ranges[name]['max']
-    pstep = st.session_state.pp_ranges[name]['step']
-    
-    c1, c2, c3, c4 = st.columns([2,1,1,1])
-    c1.write(name)
-    
-    pmin = c2.number_input(
-        "Min", min_value=0.0, max_value=pmax,
-        value=pmin, step=pstep, key=f"pp_min_{name}"
-    )
-    pmax = c3.number_input(
-        "Max", min_value=pstep, max_value=10000.0,
-        value=max(pmax, pstep), step=pstep, key=f"pp_max_{name}"
-    )
-    pstep = c4.number_input(
-        "Step", min_value=0.0,
-        max_value=(pmax - pmin) if pmax > pmin else 1.0,
-        value=pstep, step=max(1e-3, pstep/10),
-        key=f"pp_step_{name}"
-    )
-    st.session_state.pp_ranges[name].update(min=pmin, max=pmax, step=pstep)
-st.markdown("---")
-
-if st.button("Optimize WSSQ"):
-    with st.spinner("Running Optuna…"):
-        try:
-            model_print = load_prediction_model(
-                print_model_options[selected_print_model_name],
-                "printability",
-                preprocessed_input_dim,
-                len(label_encoder_print.classes_),
-            )
-            model_cell = load_prediction_model(
-                cell_model_options[selected_cell_model_name],
-                "cell_response",
-                preprocessed_input_dim,
-                len(label_encoder_cell.classes_),
-            )
-        except Exception as exc:
-            st.error(
-                "Could not load the selected prediction model. This model will only work when the "
-                "running environment matches the saved artifact dependencies. For HistGradientBoosting, "
-                "use numpy>=2.0 in the Python environment that runs Streamlit.\n\n"
-                f"Details:\n{exc}"
-            )
-            st.stop()
-
-        def objective(trial):
-            bi_vals = {
-                r['mat']: trial.suggest_float(
-                    f"bio__{r['mat']}", r['min'], r['max'], step=r['step']
-                )
-                for r in st.session_state.bio_rows
-            }
-            for m in BIOMATERIAL_OPTIONS:
-                bi_vals.setdefault(m, 0.0)
-
-            cd = 0.0 if cell_line=="NoCellCultured" else trial.suggest_float(
-                "cell_density", dr['min'], dr['max'], step=dr['step']
-            )
-
-            pp_vals = {
-                name: trial.suggest_float(
-                    f"pp__{name}",
-                    st.session_state.pp_ranges[name]['min'],
-                    st.session_state.pp_ranges[name]['max'],
-                    step=st.session_state.pp_ranges[name]['step']
-                )
-                for name in PRINT_PARAM_NAMES
-            }
-
-            feat = {**bi_vals, **pp_vals}
-            feat["Cell Density (cells/mL)"] = cd
-            feat["Cell Line"] = cell_line
-
-            X = pd.DataFrame([feat]).reindex(columns=feature_cols, fill_value=0.0)
-
-            exp_p = expected_class_value(model_print, X, preprocessor, label_encoder_print)
-            exp_c = expected_class_value(model_cell, X, preprocessor, label_encoder_cell)
-
-            np.random.seed(42)
-            # Use dynamic weights from the sidebar sliders
-            return scaffold_quality_combined(
-                exp_p, 
-                exp_c, 
-                weight_printability=w_print, 
-                weight_cell_response=w_cell
-            )
-
-        sampler = optuna.samplers.TPESampler(
-            seed=42,                    
-            n_startup_trials=30,        
-            multivariate=True,
-            group=True,                 
-            consider_prior=True
-        )
-        
-        study = optuna.create_study(
-            direction="maximize",
-            sampler=sampler,
-            pruner=optuna.pruners.MedianPruner()
-        )
-        study.optimize(objective, n_trials=int(n_trials))
-        
-        # Store results in session state to persist after rerun
-        st.session_state.best_params = study.best_trial.params
-        st.session_state.best_value = study.best_trial.value
-        st.session_state.optimized_cell_line = cell_line
-
-if 'best_params' in st.session_state:
-    st.success(f"Best WSSQ: **{st.session_state.best_value:.3f}**")
-    best_df = pd.Series(st.session_state.best_params, name="value") \
-                .rename_axis("parameter") \
-                .to_frame()
-    st.table(best_df)
-
-    st.markdown("---")
-    st.subheader("Customize Fabrication Protocol")
-    user_inquiry = st.text_area(
-        "Add specific limitations, equipment, or extra requirements:",
-        placeholder="e.g., I only have a 25G nozzle available, or I need to use a specific UV intensity of 10mW/cm²...",
-        key="user_inquiry"
-    )
-
-    if st.button("Generate Fabrication Procedure"):
-        if not gemini_key:
-            st.error("Please enter your Gemini API key in the sidebar before generating a fabrication procedure.")
-            st.stop()
-
-        with st.spinner("Generating rigorous fabrication procedure…"):
-            client = genai.Client(api_key=gemini_key)
-            
-            formatted_params = "\n".join([
-                f"- {k.replace('bio__', 'Biomaterial: ').replace('pp__', 'Print Setting: ')}: {v:.2f}"
-                for k, v in st.session_state.best_params.items()
-            ])
-            
-            # Base prompt (remains unchanged)
-            base_prompt = (
-                f"Please act as a senior tissue engineer with 15+ years of hands-on experience in 3D bioprinting for regenerative medicine. "
-                f"Write a **highly practical, bench-ready laboratory fabrication protocol** for fabricating a scaffold. "
-                f"Assume the reader is an experienced experimentalist who routinely works in a tissue engineering lab.\n\n"
-                f"**Use exactly these inputs to tailor every step:**\n"
-                f"Target Cell Line: {st.session_state.optimized_cell_line}\n"
-                f"Parameters:\n{formatted_params}\n\n"
-                f"**Critical requirements for the protocol (you MUST follow all of them):**\n"
-                f"• If Target Cell Line is 'NoCellCultured', this is **acellular 3D printing** (not bioprinting). Remove all references to cells, cell viability, cell density, and cell culturing. The final scaffold is cell-free. Change section 6 title to 'Post-Printing Incubation & Storage Instructions' and adapt its content accordingly.\n"
-                f"• If any suggested parameter is physically unrealistic (e.g. nozzle diameter 9 µm or syringe temp 2°C) or the nozzle diameter is very small relative to the cell diameter of the target cell line (when cells are used), adjust it slightly in the protocol and explicitly note the adjustment with justification.\n"
-                f"• Every quantity must be given in precise, measurable lab units (e.g., 2.5 mL, 1.2 % w/v, 10 mg/mL, 37 °C, 5 min, 150 rpm).\n"
-                f"• Include exact timings, temperatures, and workflow order to protect structural fidelity (and cell viability >85 % post-print when cells are used).\n"
-                f"• Anticipate and explicitly address common bioprinting pitfalls relevant to the given parameters (nozzle clogging, shear-induced cell death, premature gelation, filament fusion, air bubbles, etc.) and give precise mitigation steps.\n"
-                f"• Use only reagents and equipment that are standard in tissue engineering labs; if a specific brand/model is implied by the parameters, note a common equivalent.\n"
-                f"• Include simple quality-control checkpoints (visual inspection, live/dead staining timing when cells are used, etc.).\n\n"
-                f"Your response must be structured **exactly** with the following sections (no extra sections, no introductory text, no summary, no conclusions):\n"
-                f"1. Required Materials & Equipment\n"
-                f"2. Sterilization & Safety Precautions\n"
-                f"3. Bioink Preparation\n"
-                f"4. 3D Bioprinting Settings & Execution\n"
-                f"5. Post-processing & Crosslinking\n"
-                f"6. Cell Culturing & Incubation Instructions\n"
-            )
-
-            # Append user inquiry if provided
-            final_prompt = base_prompt
-            if user_inquiry:
-                final_prompt += f"\n**Additional User Constraints & Inquiries (Integrate these into the protocol):**\n{user_inquiry}"
-
-            resp = client.models.generate_content(
-                model=gemini_model,
-                contents=final_prompt,
-                config=types.GenerateContentConfig(
-                    system_instruction=(
-                        "You are a senior tissue engineer and expert experimentalist specializing in translating optimized bioprinting parameters into reproducible, high-viability laboratory protocols. "
-                        "Your protocols are used daily by PhD students and post-docs in regenerative medicine labs. "
-                        "You always prioritize: (1) maximum cell viability and function, (2) structural fidelity of the printed construct, (3) workflow efficiency under sterile conditions, and (4) safety. "
-                        "Write in clear, imperative, step-by-step language with numbered or bulleted sub-steps. "
-                        "Never be vague — give exact volumes, times, temperatures, speeds, and concentrations. "
-                        "Never add disclaimers or theoretical background unless explicitly asked."
-                    ),
-                    temperature=0.1,
-                    top_p=0.85,
-                    max_output_tokens=6144
-                )
-            )
-
-            st.markdown("## Fabrication Procedure")
-            st.markdown(resp.text)
+"""
+MLATE V3 — scaffold optimisation and protocol generation
+========================================================
+
+    streamlit run 06_webapp/app.py
+
+A thin interface over `mlate.optimize`, `mlate.wssq` and `mlate.protocol`. The
+previous release put the objective function, the WSSQ formula and the language
+model call inside this file, which meant none of them could be run, tested or
+reported without launching a browser. Everything scientific now lives in the
+package; this file collects inputs, calls it, and displays the result.
+
+Deployable to a Hugging Face Space: model artefacts are located relative to
+this file, and the only corpus data required at run time is the trimmed
+reference table that `build_app_data.py` writes beside the app.
+"""
+
+from __future__ import annotations
+
+import json
+import sys
+from pathlib import Path
+from types import SimpleNamespace
+
+import joblib
+import numpy as np
+import pandas as pd
+import streamlit as st
+
+HERE = Path(__file__).resolve().parent
+sys.path.insert(0, str(HERE.parent))
+sys.path.insert(0, str(HERE))
+
+from mlate import config as cfg          # noqa: E402
+from mlate import optimize as opt        # noqa: E402
+from mlate import protocol as proto      # noqa: E402
+from mlate import serving                # noqa: E402
+from mlate import wssq as wssq_mod       # noqa: E402
+
+from biomaterials import BIOMATERIAL_OPTIONS, BIOMATERIAL_RANGES  # noqa: E402
+from cell_lines import (ACELLULAR_TOKEN, CELL_DENSITY_RANGES,     # noqa: E402
+                        CELL_LINE_COUNTS, CELL_LINE_OPTIONS)
+from model_performance import (PERFORMANCE_GUIDE,                 # noqa: E402
+                               PERFORMANCE_GUIDE_FULL)
+
+st.set_page_config(page_title="MLATE V3", page_icon="🧬", layout="wide")
+
+# Deployment layout: a Space carries deploy/models next to the app, a checkout
+# has it at the repository root. Both are tried so the same file runs in either.
+MODEL_ROOTS = [HERE / "deploy" / "models", cfg.ROOT / "deploy" / "models"]
+PRINT_PARAMS = ["Physical Crosslinking Duration (s)",
+                "Photo Crosslinking Duration (s)",
+                "Extrusion Pressure (kPa)",
+                "Nozzle Movement Speed (mm/s)",
+                "Nozzle Diameter (µm)",
+                "Syringe Temperature (°C)",
+                "Substrate Temperature (°C)"]
+CROSSLINK = PRINT_PARAMS[:2]
+PROTOCOL_PARAMS = PRINT_PARAMS[2:]
+
+
+def models_root() -> Path | None:
+    for r in MODEL_ROOTS:
+        if (r / "classifiers").exists():
+            return r
+    return None
+
+
+# ── loading ──────────────────────────────────────────────────────────────────
+
+@st.cache_resource(show_spinner=False)
+def load_preprocessor():
+    """The release preprocessor, fitted on every row for inference."""
+    root = models_root()
+    d = root / "preprocessors"
+    return (joblib.load(d / "preprocessor.pkl"),
+            joblib.load(d / "input_columns.pkl"))
+
+
+@st.cache_resource(show_spinner=False)
+def load_manifest() -> dict:
+    root = models_root()
+    p = root / "deployment_manifest.json"
+    return json.loads(p.read_text(encoding="utf-8")) if p.exists() else {}
+
+
+@st.cache_data(show_spinner=False)
+def available_models(task: str) -> list[dict]:
+    """
+    Every model offered for one target, ranked by benchmarked weighted F1.
+
+    All three families, not only the pickled classifiers: the previous release
+    listed the conventional models alone, which meant TabICL - the strongest
+    printability model on the benchmark - was exported, reported and then not
+    offered. Discovery costs a filename parse; an estimator is opened when it
+    is chosen.
+
+    Random-split artefacts only. The application serves prediction inside the
+    design space the corpus covers - adjusting a concentration, swapping a cell
+    line, moving a pressure within observed ranges - which is the interpolation
+    regime the random protocol estimates. The study-grouped models are the
+    conservative bound for an unseen laboratory and are reported in the
+    manuscript, but they are tuned for a harder task than the one performed
+    here. The foundation models are split-independent and appear under both.
+    """
+    root = models_root()
+    if root is None:
+        return []
+    return [{"name": st_.name, "family": st_.family, "path": str(st_.path),
+             "weighted_f1": st_.weighted_f1, "cost": st_.cost_hint}
+            for st_ in serving.discover(root, task, "random")]
+
+
+def model_label(entry: dict) -> str:
+    """Menu text: the model, what kind it is, and what it scored."""
+    kind = {"ml": "", "dl": " · deep", "foundation": " · foundation"}
+    return (f"{entry['name']}{kind.get(entry['family'], '')} · "
+            f"F1 {entry['weighted_f1']:.3f}")
+
+
+@st.cache_resource(show_spinner=False)
+def load_model(path_str: str, family: str):
+    """
+    Open one artefact. Cached, because a foundation model re-supplies its
+    2,646-row context on first use and there is no reason to pay that twice.
+    """
+    return serving.load(Path(path_str), family)
+
+
+@st.cache_data(show_spinner=False)
+def load_corpus() -> tuple[pd.DataFrame | None, object]:
+    """
+    The observed formulations, shipped beside the app by `build_app_data.py`.
+
+    Needed for the two checks that a prediction alone cannot make: how far the
+    proposed formulation sits from anything published, and which real
+    formulations are closest to it. Returns (None, None) if the file is absent
+    rather than failing, so the application still optimises; the interface then
+    says the check could not be performed instead of implying it passed.
+    """
+    table, meta = HERE / "corpus_reference.parquet", HERE / "corpus_reference.json"
+    if not (table.exists() and meta.exists()):
+        return None, None
+    groups = json.loads(meta.read_text(encoding="utf-8"))
+    return pd.read_parquet(table), SimpleNamespace(**groups)
+
+
+@st.cache_data(show_spinner=False, ttl=3600)
+def list_llms() -> tuple[list[tuple], bool]:
+    """
+    The language models on offer, checked against OpenRouter's live catalogue.
+
+    Checked rather than trusted because the catalogue turns over quickly: five
+    of the eleven identifiers shipped with the previous revision had been
+    withdrawn within weeks, and a withdrawn identifier fails only when the user
+    presses Generate, after an optimisation has already been paid for. Cached
+    for an hour so the check costs one request per session.
+
+    No key is passed. The catalogue is public, and Streamlit's cache is shared
+    across every session of a deployment, so a secret used as a cache key would
+    be a secret held on behalf of all of them.
+    """
+    return proto.available_models(None)
+
+
+# ── help dialogs ───────────────────────────────────────────────────────���─────
+
+@st.dialog("Weighted Synergistic Scaffold Quality (WSSQ)", width="large")
+def show_wssq_guidance():
+    st.markdown(
+        """
+WSSQ combines **printability** and **cell response** into a single score the
+optimiser can maximise.
+
+Both components are combined by two *conjunctive* means — a weighted harmonic
+mean and a weighted geometric mean, averaged. Conjunctive means collapse toward
+zero if either component does, so neither objective can be traded away: a
+scaffold that prints perfectly but supports no cells does not score well. An
+ordinary average would allow exactly that.
+
+Two boundary rules apply. A formulation that does not extrude (printability 0)
+scores 0. An **acellular** formulation has no cell response to assess, so it is
+scored on printability alone rather than being penalised for a biological
+outcome that does not apply to it.
+
+The **cell-response weight** below is yours to set. Cell response carries more
+than biology: pore size, porosity, interconnectivity and stiffness are not known
+before fabrication and cannot be model inputs, but they strongly shape how cells
+behave, so cell response acts as a proxy for them.
+        """)
+
+
+@st.dialog("Choosing a model", width="large")
+def show_model_guidance():
+    st.markdown(
+        "Ranked by weighted F1 on the held-out test partition of the "
+        "random split (n = 530 records per target).\n\n"
+        "All three families are offered: the conventional classifiers, the six "
+        "deep networks, and the three in-context foundation models. The "
+        "highest-ranked model appears first in each menu.\n\n"
+        "**What the choice costs.** A foundation model carries the corpus "
+        "rather than fitted parameters and re-reads all 2,646 records on every "
+        "pass, so it is slower than a conventional classifier - measured at "
+        "about 1 s per batch of candidates on a GPU and about 17 s without "
+        "one, against 80 ms. Candidates are scored in batches of 32 precisely "
+        "so that this is paid once per batch rather than once per candidate; a "
+        "150-trial search is a few seconds with a conventional model, under a "
+        "minute with a foundation model on a GPU, and around three minutes on "
+        "CPU.\n\n"
+        "Models are refitted on the complete dataset for deployment. If one "
+        "fails to load because of a local package-version mismatch, choose the "
+        "next in the list.")
+    tp, tc = st.tabs(["Printability", "Cell Response"])
+    with tp:
+        st.markdown(PERFORMANCE_GUIDE["printability"])
+        with st.expander("All models"):
+            st.markdown(PERFORMANCE_GUIDE_FULL["printability"])
+    with tc:
+        st.markdown(PERFORMANCE_GUIDE["cell_response"])
+        with st.expander("All models"):
+            st.markdown(PERFORMANCE_GUIDE_FULL["cell_response"])
+
+
+@st.dialog("Optimisation trials")
+def show_trial_guidance():
+    st.markdown(
+        "Each trial is one candidate formulation, scored by the two models "
+        "and combined into WSSQ. The search is Bayesian: it models which "
+        "regions of the space produce good scores and samples there, so later "
+        "trials are better targeted than earlier ones.\n\n"
+        "The first 30 trials or so explore broadly and their scores mean "
+        "little on their own. **100 to 300 trials** is a reasonable range; "
+        "more trials help most when many variables are being searched at once.")
+
+
+@st.dialog("Getting an OpenRouter API key", width="large")
+def show_api_key_guidance():
+    st.markdown(
+        """
+Protocol generation calls a language model through **OpenRouter**, which
+provides access to models from many vendors — Anthropic, OpenAI, Google, Meta,
+DeepSeek, Mistral and others — through a single key. Nothing else in this
+application requires a key: prediction and optimisation run entirely locally.
+
+**To obtain a key**
+
+1. Go to **openrouter.ai** and create an account.
+2. Open **Keys** from the account menu and choose **Create Key**.
+3. Copy the key and paste it into the sidebar field. It is held only for this
+   browser session and is never stored or logged.
+
+**Free models.** Models whose name ends in `:free` cost nothing and need no
+credit. They are the right place to start, but they are shared and heavily
+rate-limited: a request may be refused outright when the model is busy, and the
+remedy is to wait a minute or pick another. For routine use, add a small amount
+of credit under **Credits**; a protocol typically costs well under one cent on
+the mid-range models.
+
+**Choosing a model.** The menu lists a curated selection with a note on each,
+filtered against OpenRouter's live catalogue so that a model withdrawn since
+this release does not appear. Stronger reasoning models follow the protocol's
+constraints more reliably — particularly the instruction to report an
+implausible parameter rather than quietly correct it, and the instruction not
+to invent supplier names or catalogue numbers.
+        """)
+
+
+# ── sidebar ──────────────────────────────────────────────────────────────────
+
+root = models_root()
+if root is None:
+    st.error(
+        "No model artefacts found. Expected `deploy/models/` beside this file "
+        "or at the repository root. Run `python 06_webapp/export_deployment.py` "
+        "to create them.")
+    st.stop()
+
+st.sidebar.header("Optimisation weights")
+w_cell_pct = st.sidebar.slider(
+    "Cell-response weight (%)", 0, 100, 70, 5,
+    help="Printability weight is the remainder.")
+w_print_pct = 100 - w_cell_pct
+st.sidebar.number_input("Printability weight (%)", value=w_print_pct,
+                        disabled=True)
+if st.sidebar.button("What is WSSQ?", use_container_width=True):
+    show_wssq_guidance()
+
+st.sidebar.header("Models")
+print_models = available_models("printability")
+cell_models = available_models("cell_response")
+if not print_models or not cell_models:
+    st.error("Model artefacts are present but none could be loaded.")
+    st.stop()
+
+def default_index(entries: list[dict]) -> int:
+    """
+    Which model the menu opens on.
+
+    The highest-ranked model overall when a GPU is present. On a CPU host - a
+    free Hugging Face Space, or most laptops - the highest-ranked model that is
+    not a foundation model, because an in-context model re-reads its whole
+    context on every pass and turns a search that takes seconds into one taking
+    minutes. The foundation models stay in the menu with their cost stated; the
+    difference is only what a first-time visitor is given before choosing.
+    """
+    if serving.device() == "cuda":
+        return 0
+    return next((i for i, e in enumerate(entries)
+                 if e["family"] != "foundation"), 0)
+
+
+print_entry = st.sidebar.selectbox(
+    "Printability model", print_models, format_func=model_label,
+    index=default_index(print_models))
+cell_entry = st.sidebar.selectbox(
+    "Cell-response model", cell_models, format_func=model_label,
+    index=default_index(cell_models))
+if "foundation" in (print_entry["family"], cell_entry["family"]):
+    st.sidebar.caption(
+        f"Foundation model selected - {print_entry['cost']}. These carry the "
+        f"corpus rather than fitted parameters and re-read it on every pass, "
+        f"so a search is slower than with a conventional classifier and much "
+        f"slower without a GPU.")
+if st.sidebar.button("Model performance", use_container_width=True):
+    show_model_guidance()
+
+st.sidebar.header("Search")
+n_trials = st.sidebar.number_input("Optimisation trials", 20, 2000, 150, 10)
+if st.sidebar.button("How many trials?", use_container_width=True):
+    show_trial_guidance()
+
+st.sidebar.header("Protocol generation")
+# Never prefilled from the server's own key. `type="password"` masks a value
+# on screen but still sends it to the browser, so prefilling would hand the
+# operator's key to every visitor of a public deployment. A key configured on
+# the server remains usable - `proto.api_key` falls back to it when this field
+# is empty - but it is never transmitted.
+api_key_input = st.sidebar.text_input(
+    "OpenRouter API key", type="password",
+    help="Needed only for protocol generation. Held for this browser session "
+         "only; never stored or logged.")
+if not api_key_input and proto.api_key():
+    st.sidebar.caption("A key is configured on the server; leave this blank "
+                       "to use it.")
+llm_models, llm_verified = list_llms()
+model_labels = [f"{m}  ·  {tier}" for m, _, tier, _ in llm_models]
+llm_idx = st.sidebar.selectbox(
+    "Language model", range(len(model_labels)),
+    format_func=lambda i: model_labels[i],
+    index=next((i for i, m in enumerate(llm_models)
+                if m[0] == proto.DEFAULT_MODEL), 0))
+llm_model, _vendor, _tier, llm_note = llm_models[llm_idx]
+st.sidebar.caption(
+    llm_note if llm_verified
+    else f"{llm_note}  \n_Availability unconfirmed: the OpenRouter "
+         f"catalogue could not be reached._")
+if st.sidebar.button("How to get a key", use_container_width=True):
+    show_api_key_guidance()
+
+# ── main ─────────────────────────────────────────────────────────────────────
+
+st.title("MLATE: Machine Learning Applications in Tissue Engineering")
+st.markdown(
+    "Define the ranges you can work within, and the optimiser searches them for "
+    "the formulation with the highest predicted scaffold quality. Predictions "
+    "come from models trained on 2,646 scaffold records extracted from the "
+    "literature, and are decision support rather than validated outcomes.")
+
+if "bio_rows" not in st.session_state:
+    st.session_state.bio_rows = [
+        {"mat": "Alginate (%w/v)", "min": 1.0, "max": 6.0, "step": 0.5}]
+
+st.subheader("Biomaterials")
+st.caption("Give a range for each component. The optimiser searches within it.")
+c1, c2 = st.columns([1, 5])
+if c1.button("Add biomaterial"):
+    remaining = [m for m in BIOMATERIAL_OPTIONS
+                 if m not in {r["mat"] for r in st.session_state.bio_rows}]
+    if remaining:
+        st.session_state.bio_rows.append(
+            {"mat": remaining[0], "min": 0.0, "max": 5.0, "step": 0.5})
+if c2.button("Clear all") and st.session_state.bio_rows:
+    st.session_state.bio_rows = []
+
+for i, row in enumerate(list(st.session_state.bio_rows)):
+    a, b, c, d, e = st.columns([4, 1.4, 1.4, 1.4, 0.8])
+    row["mat"] = a.selectbox("Material", BIOMATERIAL_OPTIONS,
+                             index=BIOMATERIAL_OPTIONS.index(row["mat"])
+                             if row["mat"] in BIOMATERIAL_OPTIONS else 0,
+                             key=f"mat{i}", label_visibility="collapsed")
+    obs = BIOMATERIAL_RANGES.get(row["mat"])
+    row["min"] = b.number_input("min", value=float(row["min"]), step=0.1,
+                                key=f"lo{i}")
+    row["max"] = c.number_input("max", value=float(row["max"]), step=0.1,
+                                key=f"hi{i}")
+    row["step"] = d.number_input("step", value=float(row["step"]), step=0.1,
+                                 min_value=0.01, key=f"st{i}")
+    if e.button("✕", key=f"rm{i}"):
+        st.session_state.bio_rows.pop(i)
+        st.rerun()
+    if obs:
+        note = (f"observed in corpus: {obs['min']:.3g} – {obs['max']:.3g} "
+                f"(median {obs['median']:.3g}, n = {obs['n']})")
+        if row["max"] > obs["max"] or (row["min"] > 0
+                                       and row["min"] < obs["min"]):
+            a.caption(f":orange[{note} — your range extends beyond this]")
+        else:
+            a.caption(note)
+
+st.markdown("---")
+st.subheader("Cell line and density")
+labels = [f"{c}  ({CELL_LINE_COUNTS.get(c, 0)} records)"
+          for c in CELL_LINE_OPTIONS]
+ci = st.selectbox("Cell line", range(len(CELL_LINE_OPTIONS)),
+                  format_func=lambda i: labels[i], index=0)
+cell_line = CELL_LINE_OPTIONS[ci]
+
+density_var = None
+if cell_line == ACELLULAR_TOKEN:
+    st.info(
+        "Acellular mode. Cell response does not apply, so scaffold quality is "
+        "scored on printability alone and the cell-response weight is ignored.")
+else:
+    obs = CELL_DENSITY_RANGES.get(cell_line)
+    d1, d2, d3 = st.columns(3)
+    dmin = d1.number_input("Density min (×10⁶ cells/mL)",
+                           value=float(obs["min"]) if obs else 1.0, step=0.5)
+    dmax = d2.number_input("Density max (×10⁶ cells/mL)",
+                           value=float(obs["max"]) if obs else 10.0, step=0.5)
+    dstep = d3.number_input("Density step", value=0.5, step=0.1,
+                            min_value=0.01)
+    density_var = opt.Variable(cfg.CELL_COLS[1], dmin, dmax, dstep)
+    if obs:
+        st.caption(f"observed for {cell_line}: {obs['min']:.3g} – "
+                   f"{obs['max']:.3g} (median {obs['median']:.3g})")
+
+st.markdown("---")
+st.subheader("Crosslinking")
+xl_vars = []
+for p in CROSSLINK:
+    a, b, c = st.columns(3)
+    lo = a.number_input(f"{p} — min", value=0.0, step=5.0, key=f"xlo{p}")
+    hi = b.number_input(f"{p} — max", value=300.0, step=5.0, key=f"xhi{p}")
+    stp = c.number_input("step", value=5.0, step=1.0, min_value=0.01,
+                         key=f"xst{p}")
+    xl_vars.append(opt.Variable(p, lo, hi, stp))
+
+st.subheader("Printing parameters")
+DEFAULTS = {"Extrusion Pressure (kPa)": (20.0, 200.0, 5.0),
+            "Nozzle Movement Speed (mm/s)": (1.0, 20.0, 0.5),
+            "Nozzle Diameter (µm)": (100.0, 600.0, 10.0),
+            "Syringe Temperature (°C)": (18.0, 40.0, 0.5),
+            "Substrate Temperature (°C)": (4.0, 40.0, 0.5)}
+pp_vars = []
+for p in PROTOCOL_PARAMS:
+    lo0, hi0, s0 = DEFAULTS[p]
+    a, b, c = st.columns(3)
+    lo = a.number_input(f"{p} — min", value=lo0, step=s0, key=f"plo{p}")
+    hi = b.number_input(f"{p} — max", value=hi0, step=s0, key=f"phi{p}")
+    stp = c.number_input("step", value=s0, step=0.1, min_value=0.01,
+                         key=f"pst{p}")
+    pp_vars.append(opt.Variable(p, lo, hi, stp))
+
+st.markdown("---")
+
+if st.button("Optimise scaffold quality", type="primary"):
+    if not st.session_state.bio_rows:
+        st.error("Add at least one biomaterial.")
+        st.stop()
+
+    pre, feature_columns = load_preprocessor()
+    space = opt.SearchSpace(
+        cell_line=cell_line,
+        biomaterials=[opt.Variable(r["mat"], r["min"], r["max"], r["step"])
+                      for r in st.session_state.bio_rows],
+        printing=xl_vars + pp_vars,
+        cell_density=density_var)
+
+    objective = opt.Objective(
+        space, pre, feature_columns,
+        load_model(print_entry["path"], print_entry["family"]),
+        load_model(cell_entry["path"], cell_entry["family"]),
+        print_weight=w_print_pct / 100, cell_weight=w_cell_pct / 100)
+
+    bar = st.progress(0.0, text="Searching…")
+
+    def report(done, total, best):
+        bar.progress(min(done / total, 1.0),
+                     text=f"Trial {done}/{total} · best WSSQ {best:.1f}%")
+
+    best, score, study = opt.optimise(objective, n_trials=int(n_trials),
+                                      progress=report)
+    bar.empty()
+
+    detail = objective.evaluate(best)
+    corpus, groups = load_corpus()
+    st.session_state.update(
+        best_params=best, best_value=score, best_detail=detail,
+        opt_cell_line=cell_line,
+        best_distance=(None if corpus is None
+                       else opt.distance_report(best, corpus)))
+
+if "best_params" in st.session_state:
+    best = st.session_state.best_params
+    detail = st.session_state.best_detail
+
+    st.success(f"Best WSSQ: **{st.session_state.best_value:.1f}%**")
+    m1, m2, m3 = st.columns(3)
+    m1.metric("Expected printability",
+              f"{detail['expected_printability']:.2f}", help="Scale 0–3")
+    if st.session_state.opt_cell_line != ACELLULAR_TOKEN:
+        m2.metric("Expected cell response",
+                  f"{detail['expected_cell_response']:.2f}", help="Scale 1–5")
+    m3.metric("Cell line", st.session_state.opt_cell_line)
+
+    st.subheader("Optimised formulation")
+    tidy = pd.DataFrame(
+        {"Parameter": list(best), "Value": [f"{v:.4g}" for v in best.values()]})
+    st.dataframe(tidy, use_container_width=True, hide_index=True)
+
+    st.caption(
+        "Predicted values, not measurements. The expected scores are "
+        "probability-weighted averages over the predicted class distribution, "
+        "so a value between two classes reflects genuine model uncertainty.")
+
+    dist = st.session_state.get("best_distance")
+    if dist is None:
+        st.info(
+            "The corpus reference table is not present, so how far this "
+            "formulation sits from published work could not be checked. Run "
+            "`python 06_webapp/build_app_data.py` to generate it.")
+    else:
+        d1, d2 = st.columns(2)
+        d1.metric("Parameters outside the observed range",
+                  dist["n_out_of_range"],
+                  help="Values no published study in the corpus reports. The "
+                       "models are extrapolating for these.")
+        d2.metric("Distance to nearest published formulation",
+                  f"{dist['nearest_neighbour_distance']:.3f}",
+                  help="Scaled Euclidean distance. A formulation can sit "
+                       "inside the observed range of every single variable and "
+                       "still be a combination no one has attempted; this "
+                       "number is what catches that.")
+        if dist["n_out_of_range"]:
+            with st.expander("Which parameters, and by how much"):
+                st.dataframe(pd.DataFrame([
+                    {"Parameter": k, "Proposed": f"{v['value']:.4g}",
+                     "Observed range": f"{v['observed_min']:.4g} - "
+                                       f"{v['observed_max']:.4g}"}
+                    for k, v in dist["out_of_range"].items()]),
+                    use_container_width=True, hide_index=True)
+
+    st.markdown("---")
+    st.subheader("Fabrication protocol")
+    user_inquiry = st.text_area(
+        "Constraints or equipment you must work with",
+        placeholder="e.g. only a 25 G nozzle is available; UV source is "
+                    "fixed at 10 mW/cm²",
+        key="user_inquiry")
+
+    if st.button("Generate protocol"):
+        key = proto.api_key(api_key_input)
+        if not key:
+            st.error("Enter an OpenRouter API key in the sidebar first.")
+            st.stop()
+
+        bio = {k: v for k, v in best.items() if k in BIOMATERIAL_OPTIONS}
+        printing = {k: v for k, v in best.items() if k in PRINT_PARAMS}
+        # The neighbours and the extrapolation report are what ground the
+        # prompt. Passing neither leaves the template asserting that no similar
+        # formulation exists and that the candidate is in range, and neither
+        # would have been checked.
+        corpus, groups = load_corpus()
+        neighbours = (None if corpus is None else
+                      proto.nearest_formulations(best, corpus, groups, n=3))
+        form = proto.Formulation(
+            cell_line=st.session_state.opt_cell_line,
+            biomaterials=bio, printing=printing,
+            cell_density=best.get(cfg.CELL_COLS[1]),
+            expected_printability=detail["expected_printability"],
+            expected_cell_response=detail["expected_cell_response"],
+            printability_proba=detail.get("printability_proba"),
+            cell_response_proba=detail.get("cell_response_proba"),
+            wssq=st.session_state.best_value,
+            neighbours=neighbours,
+            extrapolation=st.session_state.get("best_distance"))
+
+        # Streamed rather than awaited. A free model takes between thirty
+        # seconds and two and a half minutes to write a protocol, and a
+        # spinner held for that long is indistinguishable from a hang.
+        st.markdown("## Fabrication procedure")
+        status = st.empty()
+        stream_area = st.empty()
+        status.info(f"Generating with {llm_model}…")
+        pieces: list[str] = []
+
+        def on_chunk(piece: str) -> None:
+            pieces.append(piece)
+            stream_area.markdown("".join(pieces))
+
+        def on_retry(attempt: int, reason: str | None) -> None:
+            status.warning(
+                f"Attempt {attempt} of {proto.DEFAULT_RETRIES}: {reason}")
+
+        try:
+            out = proto.generate(
+                form, key=key, model=llm_model,
+                user_constraints=user_inquiry,
+                n_records=(len(corpus) if corpus is not None else 2646),
+                on_chunk=on_chunk, on_retry=on_retry)
+        except Exception as exc:
+            status.empty()
+            stream_area.empty()
+            st.error(f"Generation failed: {exc}")
+            st.stop()
+
+        status.empty()
+        stream_area.markdown(out["protocol"])
+        if out.get("truncated"):
+            st.warning(
+                "The model reached its output limit and the protocol is "
+                "incomplete. Generate again, or choose another model.")
+        with st.expander("Generation record"):
+            st.json({k: v for k, v in out.items() if k != "protocol"})
+        st.download_button("Download protocol (Markdown)",
+                           out["protocol"],
+                           file_name="mlate_protocol.md")
+
+st.markdown("---")
+st.caption(
+    "MLATE V3 — decision support for 3D-printed and bioprinted scaffolds. "
+    "Predictions narrow the experimental search space; they do not replace "
+    "experimental validation.")

biomaterials.py

@@ -1,124 +1,926 @@
+"""
+Biomaterial vocabulary and observed concentration ranges.
+
+GENERATED by 06_webapp/build_app_data.py - do not edit by hand.
+Regenerate after any change to the dataset or the benchmark tables.
+"""
+
 BIOMATERIAL_OPTIONS = [
-    "Alginate (%w/v)",
-    "PVA-HA (%w/v)",
-    "CaSO4 (%w/v)",
-    "Na2HPO4 (%w/v)",
-    "Gelatin (%w/v)",
-    "GelMA (%w/v)",
-    "laponite (%w/v)",
-    "graphene oxide (%w/v)",
-    "hydroxyapatite (%w/v)",
-    "Hyaluronic_Acid (%w/v)",
-    "hyaluronan metacrylate (%w/v)",
-    "NorHA (%w/v)",
-    "Fibroin/Fibrinogen (%w/v)",
-    "Pluronic P-123 (%w/v)",
-    "Collagen (%w/v)",
-    "Chitosan (%w/v)",
-    "CS-AEMA (%w/v)",
-    "RGD (mM)",
-    "TCP (%w/v)",
-    "Gellan (%w/v)",
-    "bioactive glass (%w/v)",
-    "Nano/Methycellulose (%w/v)",
-    "PEGTA (%w/v)",
-    "PEGMA (%w/v)",
-    "PEGDA (%w/v)",
-    "Agarose (%w/v)",
-    " hyaluronic acid+ Ph moieties (%w/v)",
-    "matrigel (%w/v)",
-    "CaCl2(mM)",
-    "NaCl(mM)",
-    "BaCl2(mM)",
-    "SrCl2(mM)",
-    "CaCO3 (mM)",
-    "Genipin (%w/v)",
-    "PVA (%wt)",
-    "trans-glutaminase (%w/v)",
-    "alginate lyase (U/ml)",
-    "D-glucose (%w/v)",
-    "PLGA (%w/v)",
-    "vascular tissued-derived dECM (%w/v)",
-    "PEG-8-SH (mM)",
-    "Alginate dialdehyde (%w/v)",
-    "Alginate sulfate (%w/v)",
-    "RGD-modified alginate (%w/v)",
-    "poly(N-isopropylacrylamide) grafted hyaluronan (%w/v)",
-    "chondroitin sulfate methacrylate (%w/v)",
-    "PCL (%w/v)",
-    "alginate methacrylate (%w/v)",
-    "HRP (U/ml)",
-    "Pluronic F127 (%w/v)/Lutrol F127 (%w/v)",
-    "Irgacure 2959 (%w/v)",
-    "Eosin Y (%w/v)",
-    "Ruthenium (mM)",
-    "sodium persulfate (SPS) (mM)",
-    "HEPES (mM)",
-    "LAP (%w/v)",
-    "glutaraldehyde (%w/v)",
-    "PBS  (M)",
-    "glycerol (%w/v)",
-    "cECM (%w/v)",
-    "gel-fu(%w/v)",
-    "Rose Bengal (%w/v)",
-    "Vitamin B2(%w/v)",
-    "VEGF(%w/v)",
-    "Polypyrrole:PSS(%w/v)",
-    "boratebioactiveglass(%w/v)",
-    "astaxanthin(%w/v)",
-    "PRP (%v/v)",
-    "methacrylated collagen (%w/v)",
-    "α-Toc (µM)",
-    "ascorbic acid (mM)",
-    "Liver dECM(%w/v)",
-    "galactosylated alginate (%w/v)",
-    "SC-PEG(%w/v)",
-    "SFMA-L(%w/v)",
-    "SFMA-M(%w/v)",
-    "SFMA-H(%w/v)",
-    "KdECMMA(%w/v)",
-    "BA silk fibronin (%w/v)",
-    "Carrageenan(%v)",
-    "Carbopol ETD 2020 NF (%w/v)",
-    "Carbopol  Ultrez 10 NF(%w/v)",
-    "Carbopol  NF-980(%w/v)",
-    "FBS (%v/v)",
-    "MeTro (%w/v)",
-    "Triethanolamine (%v/v)",
-    "PEG-Fibrinogen (%w/v)",
-    "polyethylene glycol dimethacrylate (%w/v)",
-    "aprotinin (µg/ml)",
-    "gold nanorod (mg/mL)",
-    "egg white (w/v)",
-    "1-Vinyl-2-Pyrrolidione (v/v)",
-    "carboxyl functionalized carbon nanotubes  (%w/v)",
-    "polyHIPE  (%w/v)",
-    "β-D galactose (mM)",
-    "hydrogen peroxide (H2O2) (%v/v)",
-    "lactic acid v/v",
-    "NorCol (%w/v)",
-    "DDT (%w/v)",
-    "ammonium persulfate (mM)",
-    "diTyr-RGD (mM)",
-    "PHEG-Tyr (%w/v)",
-    "MMP2-degradable peptide (%w/v)",
-    "KdECM (%w/v)",
-    "EDC (mg)",
-    "NHS (mg)",
-    "VA086 (%w/v)",
-    "PGS (%w/v)",
-    "thiolated HA (%w/v)",
-    "boron nitride nanotubes (%w/v)",
-    "PEDOT:PSS (ul)",
-    "KCl (mM)",
-    "skeletal muscle ECM methacrylate (%w/v)",
-    "PEO (%w/v)",
-    "Carbon dots (mg/ml)",
-    "Laminin (ug/ml)",
-    "DF-PEG (%w/v)",
-    "omenta ECM (%w/v)",
-    "thrombin (unit/ml)",
-    "Carbon nanotube (CNT) (w/v)",
-    "Phytagel(%v)",
-    "Laponite-XLG  (%w/w)"
-]
+    'Alginate (%w/v)',
+    'PVA-HA (%w/v)',
+    'CaSO4 (%w/v)',
+    'Na2HPO4 (%w/v)',
+    'Gelatin (%w/v)',
+    'GelMA (%w/v)',
+    'laponite (%w/v)',
+    'graphene oxide (%w/v)',
+    'hydroxyapatite (%w/v)',
+    'Hyaluronic_Acid (%w/v)',
+    'hyaluronan methacrylate (%w/v)',
+    'NorHA (%w/v)',
+    'Fibroin/Fibrinogen (%w/v)',
+    'Pluronic P-123 (%w/v)',
+    'Collagen (%w/v)',
+    'Chitosan (%w/v)',
+    'CS-AEMA (%w/v)',
+    'RGD (mM)',
+    'TCP (%w/v)',
+    'Gellan (%w/v)',
+    'bioactive glass (%w/v)',
+    'Nano/Methylcellulose (%w/v)',
+    'PEGTA (%w/v)',
+    'PEGMA (%w/v)',
+    'PEGDA (%w/v)',
+    'Agarose (%w/v)',
+    'hyaluronic acid+ Ph moieties (%w/v)',
+    'matrigel (%w/v)',
+    'CaCl2 (mM)',
+    'NaCl (mM)',
+    'BaCl2 (mM)',
+    'SrCl2 (mM)',
+    'CaCO3 (mM)',
+    'Genipin (%w/v)',
+    'PVA (%wt)',
+    'trans-glutaminase (%w/v)',
+    'alginate lyase (U/ml)',
+    'D-glucose (%w/v)',
+    'PLGA (%w/v)',
+    'vascular tissue-derived dECM (%w/v)',
+    'PEG-8-SH (mM)',
+    'Alginate dialdehyde (%w/v)',
+    'Alginate sulfate (%w/v)',
+    'RGD-modified alginate (%w/v)',
+    'poly(N-isopropylacrylamide) grafted hyaluronan (%w/v)',
+    'chondroitin sulfate methacrylate (%w/v)',
+    'PCL (%w/v)',
+    'alginate methacrylate (%w/v)',
+    'HRP (U/ml)',
+    'Pluronic F127 / Lutrol F127 (%w/v)',
+    'Irgacure 2959 (%w/v)',
+    'Eosin Y (%w/v)',
+    'Ruthenium (mM)',
+    'sodium persulfate (SPS) (mM)',
+    'HEPES (mM)',
+    'LAP (%w/v)',
+    'glutaraldehyde (%w/v)',
+    'PBS (M)',
+    'glycerol (%w/v)',
+    'cECM (%w/v)',
+    'gel-fu (%w/v)',
+    'Rose Bengal (%w/v)',
+    'Vitamin B2 (%w/v)',
+    'VEGF (%w/v)',
+    'Polypyrrole:PSS (%w/v)',
+    'borate bioactive glass (%w/v)',
+    'astaxanthin (%w/v)',
+    'PRP (%v/v)',
+    'methacrylated collagen (%w/v)',
+    'α-Toc (µM)',
+    'ascorbic acid (mM)',
+    'Liver dECM (%w/v)',
+    'galactosylated alginate (%w/v)',
+    'SC-PEG (%w/v)',
+    'SFMA-L (%w/v)',
+    'SFMA-M (%w/v)',
+    'SFMA-H (%w/v)',
+    'KdECMMA (%w/v)',
+    'BA silk fibroin (%w/v)',
+    'Carrageenan (%v)',
+    'Carbopol ETD 2020 NF (%w/v)',
+    'Carbopol Ultrez 10 NF (%w/v)',
+    'Carbopol NF-980 (%w/v)',
+    'FBS (%v/v)',
+    'MeTro (%w/v)',
+    'Triethanolamine (%v/v)',
+    'PEG-Fibrinogen (%w/v)',
+    'polyethylene glycol dimethacrylate (%w/v)',
+    'aprotinin (µg/ml)',
+    'gold nanorod (mg/mL)',
+    'egg white (w/v)',
+    '1-Vinyl-2-pyrrolidone (v/v)',
+    'carboxyl functionalized carbon nanotubes (%w/v)',
+    'polyHIPE (%w/v)',
+    'β-D galactose (mM)',
+    'hydrogen peroxide (H2O2) (%v/v)',
+    'lactic acid (v/v)',
+    'NorCol (%w/v)',
+    'DTT (%w/v)',
+    'ammonium persulfate (mM)',
+    'diTyr-RGD (mM)',
+    'PHEG-Tyr (%w/v)',
+    'MMP2-degradable peptide (%w/v)',
+    'KdECM (%w/v)',
+    'EDC (mg)',
+    'NHS (mg)',
+    'VA086 (%w/v)',
+    'PGS (%w/v)',
+    'thiolated HA (%w/v)',
+    'boron nitride nanotubes (%w/v)',
+    'PEDOT:PSS (ul)',
+    'KCl (mM)',
+    'skeletal muscle ECM methacrylate (%w/v)',
+    'PEO (%w/v)',
+    'Carbon dots (mg/ml)',
+    'Laminin (ug/ml)',
+    'DF-PEG (%w/v)',
+    'omentum ECM (%w/v)',
+    'thrombin (unit/ml)',
+    'Carbon nanotube (CNT) (w/v)',
+    'Phytagel (%v)',
+    'Laponite-XLG (%w/w)',
+    'sodium carboxymethyl cellulose (mg)',
+    'lysozyme amyloid nanofibrils:gold nanoparticles (mg)',
+    'Lysozyme amyloid nanofibrils (mg)',
+    'rGO (mg/ml)',
+    'Methacrylated gellan gum (%w/v)',
+    'Acetylsalicylic Acid (%w/w)',
+    'PVA methacrylate (%w/v)',
+    'MXene (mg/ml)',
+]
+
+# Observed range among formulations that CONTAIN each material.
+# The UI seeds its range widgets from these so a user starts
+# inside the region the models were trained on; values outside
+# are permitted but flagged as extrapolation.
+BIOMATERIAL_RANGES = {
+    "Alginate (%w/v)": {
+        "min": 0.25,
+        "max": 20.0,
+        "median": 4.0,
+        "n": 1138
+    },
+    "PVA-HA (%w/v)": {
+        "min": 0.3,
+        "max": 2.5,
+        "median": 2.0,
+        "n": 15
+    },
+    "CaSO4 (%w/v)": {
+        "min": 0.03,
+        "max": 25.71,
+        "median": 1.0,
+        "n": 32
+    },
+    "Na2HPO4 (%w/v)": {
+        "min": 0.12,
+        "max": 0.15,
+        "median": 0.15,
+        "n": 9
+    },
+    "Gelatin (%w/v)": {
+        "min": 0.5,
+        "max": 80.0,
+        "median": 5.0,
+        "n": 797
+    },
+    "GelMA (%w/v)": {
+        "min": 1.0,
+        "max": 30.0,
+        "median": 7.0,
+        "n": 880
+    },
+    "laponite (%w/v)": {
+        "min": 0.05,
+        "max": 2.3,
+        "median": 2.3,
+        "n": 96
+    },
+    "graphene oxide (%w/v)": {
+        "min": 0.001,
+        "max": 1.5,
+        "median": 0.5,
+        "n": 43
+    },
+    "hydroxyapatite (%w/v)": {
+        "min": 2.0,
+        "max": 70.0,
+        "median": 70.0,
+        "n": 43
+    },
+    "Hyaluronic_Acid (%w/v)": {
+        "min": 0.2,
+        "max": 2.0,
+        "median": 0.3,
+        "n": 62
+    },
+    "hyaluronan methacrylate (%w/v)": {
+        "min": 0.1,
+        "max": 6.0,
+        "median": 2.0,
+        "n": 151
+    },
+    "NorHA (%w/v)": {
+        "min": 2.0,
+        "max": 2.0,
+        "median": 2.0,
+        "n": 4
+    },
+    "Fibroin/Fibrinogen (%w/v)": {
+        "min": 0.009375,
+        "max": 25.0,
+        "median": 2.0,
+        "n": 207
+    },
+    "Pluronic P-123 (%w/v)": {
+        "min": 40.0,
+        "max": 60.0,
+        "median": 50.0,
+        "n": 12
+    },
+    "Collagen (%w/v)": {
+        "min": 0.01,
+        "max": 7.8,
+        "median": 2.0,
+        "n": 85
+    },
+    "Chitosan (%w/v)": {
+        "min": 1.0,
+        "max": 62.0,
+        "median": 2.0,
+        "n": 74
+    },
+    "CS-AEMA (%w/v)": {
+        "min": 4.0,
+        "max": 4.0,
+        "median": 4.0,
+        "n": 8
+    },
+    "RGD (mM)": {
+        "min": 3.0,
+        "max": 3.0,
+        "median": 3.0,
+        "n": 1
+    },
+    "TCP (%w/v)": {
+        "min": 0.5,
+        "max": 3.0,
+        "median": 0.5,
+        "n": 9
+    },
+    "Gellan (%w/v)": {
+        "min": 0.5,
+        "max": 150.0,
+        "median": 150.0,
+        "n": 54
+    },
+    "bioactive glass (%w/v)": {
+        "min": 1.0,
+        "max": 50.0,
+        "median": 1.0,
+        "n": 7
+    },
+    "Nano/Methylcellulose (%w/v)": {
+        "min": 0.25,
+        "max": 80.0,
+        "median": 4.0,
+        "n": 157
+    },
+    "PEGTA (%w/v)": {
+        "min": 1.0,
+        "max": 3.0,
+        "median": 2.0,
+        "n": 33
+    },
+    "PEGMA (%w/v)": {
+        "min": 12.5,
+        "max": 100.0,
+        "median": 17.5,
+        "n": 9
+    },
+    "PEGDA (%w/v)": {
+        "min": 1.0,
+        "max": 20.0,
+        "median": 8.25,
+        "n": 119
+    },
+    "Agarose (%w/v)": {
+        "min": 0.5,
+        "max": 60.0,
+        "median": 24.0,
+        "n": 51
+    },
+    "hyaluronic acid+ Ph moieties (%w/v)": {
+        "min": 0.1,
+        "max": 1.5,
+        "median": 0.8,
+        "n": 8
+    },
+    "matrigel (%w/v)": {
+        "min": 5.0,
+        "max": 50.0,
+        "median": 10.0,
+        "n": 19
+    },
+    "CaCl2 (mM)": {
+        "min": 0.001,
+        "max": 1500.0,
+        "median": 100.0,
+        "n": 1057
+    },
+    "NaCl (mM)": {
+        "min": 0.72,
+        "max": 350.0,
+        "median": 145.0,
+        "n": 23
+    },
+    "BaCl2 (mM)": {
+        "min": 55.0,
+        "max": 60.0,
+        "median": 55.0,
+        "n": 12
+    },
+    "SrCl2 (mM)": {
+        "min": 20.0,
+        "max": 70.0,
+        "median": 20.0,
+        "n": 6
+    },
+    "CaCO3 (mM)": {
+        "min": 15.88,
+        "max": 25.71,
+        "median": 18.0,
+        "n": 9
+    },
+    "Genipin (%w/v)": {
+        "min": 0.025,
+        "max": 1.0,
+        "median": 0.025,
+        "n": 18
+    },
+    "PVA (%wt)": {
+        "min": 3.0,
+        "max": 15.0,
+        "median": 15.0,
+        "n": 50
+    },
+    "trans-glutaminase (%w/v)": {
+        "min": 0.04,
+        "max": 6.0,
+        "median": 1.0,
+        "n": 63
+    },
+    "alginate lyase (U/ml)": {
+        "min": 0.5,
+        "max": 500.0,
+        "median": 5.0,
+        "n": 7
+    },
+    "D-glucose (%w/v)": {
+        "min": 4.4,
+        "max": 4.4,
+        "median": 4.4,
+        "n": 5
+    },
+    "PLGA (%w/v)": {
+        "min": 100.0,
+        "max": 100.0,
+        "median": 100.0,
+        "n": 4
+    },
+    "vascular tissue-derived dECM (%w/v)": {
+        "min": 1.0,
+        "max": 3.0,
+        "median": 3.0,
+        "n": 7
+    },
+    "PEG-8-SH (mM)": {
+        "min": 2.25,
+        "max": 8.0,
+        "median": 2.25,
+        "n": 10
+    },
+    "Alginate dialdehyde (%w/v)": {
+        "min": 2.0,
+        "max": 7.5,
+        "median": 3.75,
+        "n": 48
+    },
+    "Alginate sulfate (%w/v)": {
+        "min": 1.0,
+        "max": 1.0,
+        "median": 1.0,
+        "n": 26
+    },
+    "RGD-modified alginate (%w/v)": {
+        "min": 1.0,
+        "max": 1.0,
+        "median": 1.0,
+        "n": 2
+    },
+    "poly(N-isopropylacrylamide) grafted hyaluronan (%w/v)": {
+        "min": 15.0,
+        "max": 15.0,
+        "median": 15.0,
+        "n": 3
+    },
+    "chondroitin sulfate methacrylate (%w/v)": {
+        "min": 5.0,
+        "max": 5.0,
+        "median": 5.0,
+        "n": 1
+    },
+    "PCL (%w/v)": {
+        "min": 1.0,
+        "max": 100.0,
+        "median": 8.0,
+        "n": 34
+    },
+    "alginate methacrylate (%w/v)": {
+        "min": 1.0,
+        "max": 3.0,
+        "median": 3.0,
+        "n": 24
+    },
+    "HRP (U/ml)": {
+        "min": 5.0,
+        "max": 100.0,
+        "median": 15.0,
+        "n": 68
+    },
+    "Pluronic F127 / Lutrol F127 (%w/v)": {
+        "min": 3.0,
+        "max": 100.0,
+        "median": 6.0,
+        "n": 83
+    },
+    "Irgacure 2959 (%w/v)": {
+        "min": 0.01,
+        "max": 2.0,
+        "median": 0.25,
+        "n": 530
+    },
+    "Eosin Y (%w/v)": {
+        "min": 0.5,
+        "max": 100.0,
+        "median": 0.5,
+        "n": 32
+    },
+    "Ruthenium (mM)": {
+        "min": 0.254,
+        "max": 1.0,
+        "median": 0.5,
+        "n": 28
+    },
+    "sodium persulfate (SPS) (mM)": {
+        "min": 2.52,
+        "max": 10.0,
+        "median": 5.0,
+        "n": 28
+    },
+    "HEPES (mM)": {
+        "min": 10.0,
+        "max": 25.0,
+        "median": 10.0,
+        "n": 22
+    },
+    "LAP (%w/v)": {
+        "min": 0.01,
+        "max": 4.46,
+        "median": 0.4,
+        "n": 310
+    },
+    "glutaraldehyde (%w/v)": {
+        "min": 0.125,
+        "max": 0.4,
+        "median": 0.25,
+        "n": 62
+    },
+    "PBS (M)": {
+        "min": 0.082,
+        "max": 0.328,
+        "median": 0.165,
+        "n": 3
+    },
+    "glycerol (%w/v)": {
+        "min": 10.0,
+        "max": 10.0,
+        "median": 10.0,
+        "n": 21
+    },
+    "cECM (%w/v)": {
+        "min": 0.1,
+        "max": 20.0,
+        "median": 4.6,
+        "n": 118
+    },
+    "gel-fu (%w/v)": {
+        "min": 10.0,
+        "max": 155.0,
+        "median": 100.0,
+        "n": 15
+    },
+    "Rose Bengal (%w/v)": {
+        "min": 1.0,
+        "max": 5.0,
+        "median": 5.0,
+        "n": 14
+    },
+    "Vitamin B2 (%w/v)": {
+        "min": 0.02,
+        "max": 0.2,
+        "median": 0.02,
+        "n": 20
+    },
+    "VEGF (%w/v)": {
+        "min": 0.01,
+        "max": 10.0,
+        "median": 1.0,
+        "n": 6
+    },
+    "Polypyrrole:PSS (%w/v)": {
+        "min": 0.1,
+        "max": 0.4,
+        "median": 0.2,
+        "n": 3
+    },
+    "borate bioactive glass (%w/v)": {
+        "min": 0.1,
+        "max": 0.1,
+        "median": 0.1,
+        "n": 1
+    },
+    "astaxanthin (%w/v)": {
+        "min": 0.01,
+        "max": 0.01,
+        "median": 0.01,
+        "n": 1
+    },
+    "PRP (%v/v)": {
+        "min": 20.0,
+        "max": 20.0,
+        "median": 20.0,
+        "n": 3
+    },
+    "methacrylated collagen (%w/v)": {
+        "min": 0.2,
+        "max": 50.0,
+        "median": 0.48,
+        "n": 25
+    },
+    "\u03b1-Toc (\u00b5M)": {
+        "min": 100.0,
+        "max": 100.0,
+        "median": 100.0,
+        "n": 5
+    },
+    "ascorbic acid (mM)": {
+        "min": 3.4,
+        "max": 3.4,
+        "median": 3.4,
+        "n": 5
+    },
+    "Liver dECM (%w/v)": {
+        "min": 0.5,
+        "max": 100.0,
+        "median": 2.0,
+        "n": 139
+    },
+    "galactosylated alginate (%w/v)": {
+        "min": 0.375,
+        "max": 1.0,
+        "median": 1.0,
+        "n": 22
+    },
+    "SC-PEG (%w/v)": {
+        "min": 1.44,
+        "max": 1.44,
+        "median": 1.44,
+        "n": 4
+    },
+    "SFMA-L (%w/v)": {
+        "min": 10.0,
+        "max": 10.0,
+        "median": 10.0,
+        "n": 1
+    },
+    "SFMA-M (%w/v)": {
+        "min": 10.0,
+        "max": 10.0,
+        "median": 10.0,
+        "n": 1
+    },
+    "SFMA-H (%w/v)": {
+        "min": 10.0,
+        "max": 10.0,
+        "median": 10.0,
+        "n": 1
+    },
+    "KdECMMA (%w/v)": {
+        "min": 1.0,
+        "max": 3.0,
+        "median": 2.0,
+        "n": 15
+    },
+    "BA silk fibroin (%w/v)": {
+        "min": 0.5,
+        "max": 3.0,
+        "median": 1.5,
+        "n": 32
+    },
+    "Carrageenan (%v)": {
+        "min": 0.5,
+        "max": 1.5,
+        "median": 1.0,
+        "n": 43
+    },
+    "Carbopol ETD 2020 NF (%w/v)": {
+        "min": 0.1,
+        "max": 1.2,
+        "median": 0.5,
+        "n": 32
+    },
+    "Carbopol Ultrez 10 NF (%w/v)": {
+        "min": 1.5,
+        "max": 1.5,
+        "median": 1.5,
+        "n": 6
+    },
+    "Carbopol NF-980 (%w/v)": {
+        "min": 1.2,
+        "max": 1.2,
+        "median": 1.2,
+        "n": 4
+    },
+    "FBS (%v/v)": {
+        "min": 10.0,
+        "max": 10.0,
+        "median": 10.0,
+        "n": 34
+    },
+    "MeTro (%w/v)": {
+        "min": 7.5,
+        "max": 7.5,
+        "median": 7.5,
+        "n": 28
+    },
+    "Triethanolamine (%v/v)": {
+        "min": 3.0,
+        "max": 3.0,
+        "median": 3.0,
+        "n": 4
+    },
+    "PEG-Fibrinogen (%w/v)": {
+        "min": 1.0,
+        "max": 1.0,
+        "median": 1.0,
+        "n": 8
+    },
+    "polyethylene glycol dimethacrylate (%w/v)": {
+        "min": 1.0,
+        "max": 1.0,
+        "median": 1.0,
+        "n": 50
+    },
+    "aprotinin (\u00b5g/ml)": {
+        "min": 0.2,
+        "max": 10.0,
+        "median": 0.2,
+        "n": 3
+    },
+    "gold nanorod (mg/mL)": {
+        "min": 0.1,
+        "max": 0.1,
+        "median": 0.1,
+        "n": 24
+    },
+    "egg white (w/v)": {
+        "min": 1.0,
+        "max": 3.0,
+        "median": 2.0,
+        "n": 10
+    },
+    "1-Vinyl-2-pyrrolidone (v/v)": {
+        "min": 0.75,
+        "max": 0.75,
+        "median": 0.75,
+        "n": 4
+    },
+    "carboxyl functionalized carbon nanotubes (%w/v)": {
+        "min": 0.3,
+        "max": 2.0,
+        "median": 1.0,
+        "n": 9
+    },
+    "polyHIPE (%w/v)": {
+        "min": 1.0,
+        "max": 5.0,
+        "median": 3.0,
+        "n": 5
+    },
+    "\u03b2-D galactose (mM)": {
+        "min": 40.0,
+        "max": 40.0,
+        "median": 40.0,
+        "n": 2
+    },
+    "hydrogen peroxide (H2O2) (%v/v)": {
+        "min": 0.09,
+        "max": 3.87,
+        "median": 3.87,
+        "n": 55
+    },
+    "lactic acid (v/v)": {
+        "min": 3.0,
+        "max": 3.0,
+        "median": 3.0,
+        "n": 30
+    },
+    "NorCol (%w/v)": {
+        "min": 0.2,
+        "max": 1.0,
+        "median": 0.6,
+        "n": 13
+    },
+    "DTT (%w/v)": {
+        "min": 0.5,
+        "max": 8.0,
+        "median": 0.7,
+        "n": 10
+    },
+    "ammonium persulfate (mM)": {
+        "min": 3.75,
+        "max": 5.0,
+        "median": 4.375,
+        "n": 16
+    },
+    "diTyr-RGD (mM)": {
+        "min": 2.0,
+        "max": 6.0,
+        "median": 4.0,
+        "n": 8
+    },
+    "PHEG-Tyr (%w/v)": {
+        "min": 10.0,
+        "max": 10.0,
+        "median": 10.0,
+        "n": 12
+    },
+    "MMP2-degradable peptide (%w/v)": {
+        "min": 0.1,
+        "max": 0.5,
+        "median": 0.3,
+        "n": 8
+    },
+    "KdECM (%w/v)": {
+        "min": 1.0,
+        "max": 5.0,
+        "median": 3.3,
+        "n": 11
+    },
+    "EDC (mg)": {
+        "min": 50.0,
+        "max": 50.0,
+        "median": 50.0,
+        "n": 22
+    },
+    "NHS (mg)": {
+        "min": 10.0,
+        "max": 25.0,
+        "median": 10.0,
+        "n": 22
+    },
+    "VA086 (%w/v)": {
+        "min": 0.5,
+        "max": 20.0,
+        "median": 0.75,
+        "n": 10
+    },
+    "PGS (%w/v)": {
+        "min": 2.0,
+        "max": 20.0,
+        "median": 11.0,
+        "n": 2
+    },
+    "thiolated HA (%w/v)": {
+        "min": 0.04,
+        "max": 0.067,
+        "median": 0.05,
+        "n": 10
+    },
+    "boron nitride nanotubes (%w/v)": {
+        "min": 0.05,
+        "max": 0.1,
+        "median": 0.075,
+        "n": 3
+    },
+    "PEDOT:PSS (ul)": {
+        "min": 0.1,
+        "max": 0.3,
+        "median": 0.1,
+        "n": 18
+    },
+    "KCl (mM)": {
+        "min": 5.0,
+        "max": 5.0,
+        "median": 5.0,
+        "n": 15
+    },
+    "skeletal muscle ECM methacrylate (%w/v)": {
+        "min": 3.0,
+        "max": 3.0,
+        "median": 3.0,
+        "n": 7
+    },
+    "PEO (%w/v)": {
+        "min": 0.4,
+        "max": 0.4,
+        "median": 0.4,
+        "n": 9
+    },
+    "Carbon dots (mg/ml)": {
+        "min": 20.0,
+        "max": 20.0,
+        "median": 20.0,
+        "n": 9
+    },
+    "Laminin (ug/ml)": {
+        "min": 10.0,
+        "max": 93.75,
+        "median": 70.0,
+        "n": 13
+    },
+    "DF-PEG (%w/v)": {
+        "min": 25.0,
+        "max": 25.0,
+        "median": 25.0,
+        "n": 6
+    },
+    "omentum ECM (%w/v)": {
+        "min": 2.0,
+        "max": 2.0,
+        "median": 2.0,
+        "n": 2
+    },
+    "thrombin (unit/ml)": {
+        "min": 2.5,
+        "max": 50.0,
+        "median": 10.0,
+        "n": 41
+    },
+    "Carbon nanotube (CNT) (w/v)": {
+        "min": 0.5,
+        "max": 6.0,
+        "median": 1.5,
+        "n": 18
+    },
+    "Phytagel (%v)": {
+        "min": 0.3,
+        "max": 0.3,
+        "median": 0.3,
+        "n": 1
+    },
+    "Laponite-XLG (%w/w)": {
+        "min": 2.3,
+        "max": 2.3,
+        "median": 2.3,
+        "n": 25
+    },
+    "sodium carboxymethyl cellulose (mg)": {
+        "min": 75.0,
+        "max": 150.0,
+        "median": 100.0,
+        "n": 60
+    },
+    "lysozyme amyloid nanofibrils:gold nanoparticles (mg)": {
+        "min": 2.5,
+        "max": 37.5,
+        "median": 12.5,
+        "n": 6
+    },
+    "Lysozyme amyloid nanofibrils (mg)": {
+        "min": 2.5,
+        "max": 37.5,
+        "median": 12.5,
+        "n": 31
+    },
+    "rGO (mg/ml)": {
+        "min": 150.0,
+        "max": 150.0,
+        "median": 150.0,
+        "n": 2
+    },
+    "Methacrylated gellan gum (%w/v)": {
+        "min": 2.0,
+        "max": 3.5,
+        "median": 3.0,
+        "n": 15
+    },
+    "Acetylsalicylic Acid (%w/w)": {
+        "min": 10.0,
+        "max": 10.0,
+        "median": 10.0,
+        "n": 3
+    },
+    "PVA methacrylate (%w/v)": {
+        "min": 1.0,
+        "max": 5.0,
+        "median": 5.0,
+        "n": 11
+    },
+    "MXene (mg/ml)": {
+        "min": 0.1,
+        "max": 5.0,
+        "median": 0.75,
+        "n": 6
+    }
+}

cell_lines.py

@@ -1,178 +1,1320 @@
+"""
+Cell-line vocabulary and observed seeding densities.
+
+GENERATED by 06_webapp/build_app_data.py - do not edit by hand.
+Regenerate after any change to the dataset or the benchmark tables.
+"""
+
+# Ordered by frequency in the corpus, with the acellular token
+# first because it selects a different mode: cell response is
+# not applicable and WSSQ falls back to printability alone.
+ACELLULAR_TOKEN = 'NoCellCultured'
+
 CELL_LINE_OPTIONS = [
-    "NoCellCultured",
-    "chondrocyteyte",
-    "HepG2",
-    "bMSCs",
-    "HUVECs",
-    "NIH3T3",
-    "MESCs",
-    "hiPSC-CMs /ATCCs",
-    "CPCs",
-    "L929",
-    "Myoblast cells",
-    "hiPSCs",
-    "HepaRG",
-    "hESCs",
-    "10T1/2",
-    "Cardiac progenitor cells",
-    "NSCLC PDX",
-    "RAMECs",
-    "hASCs",
-    "HAVIC",
-    "Primary mouse hepatocyte",
-    "PTECs",
-    "human nasoseptal chondrocytes",
-    "PDX",
-    "HPFs",
-    "U87-MG",
-    "ESCs",
-    "HASSMC",
-    "dermal fibroblasts",
-    "MC3T3-E1",
-    "Schwann cells",
-    "hiPSC-CMs and HS-27A",
-    "Saos-2 ",
-    "SU3",
-    "hTMSCs",
-    "HACs",
-    "HADMSCs",
-    "HeLa",
-    "human primary kidney cells",
-    "myoblasts",
-    "MSCs",
-    "human primary kidneycells",
-    "293FT",
-    "HEK 293FT",
-    "Wnt3a-293FT",
-    "RSC96/HUVECs",
-    "human adipogenic mesenchymal stem cells",
-    "HEPG2/ECs",
-    "HUVECs/MSCs",
-    "RHECs",
-    "Human non-small cell lung cancer line Calu-3 (Calu-3)",
-    "HL-1",
-    "mouse cardiac cells",
-    "IMR-90",
-    "EPCs",
-    "MRC5",
-    "rMSC",
-    "basil plant cell",
-    "hNCs",
-    "A549",
-    "human induced pluripotent stem cell-derived cardiomyocytes",
-    "bMSCs/hACs",
-    "EA.hy 926 cells",
-    "HepG2/C3A",
-    "human epithelial lung carcinoma cells",
-    "Human cardiac fibroblasts",
-    "hTERT-MSC",
-    "cardiomyocytes",
-    "Huh7",
-    "NRCMs",
-    "HCF",
-    "Wnt reporter-293FT",
-    "neonatal rat ventricular CFs",
-    "human coronary artery endothelial cells",
-    "hiPSC-CM / fibroblasts",
-    "primary mouse hepatocyte",
-    "NIH3T3/ HUVECs",
-    "murine macrophage-like cell line",
-    "Endothelial cells",
-    "Human aortic VIC",
-    "sADSC",
-    "HUVECs/H9C2",
-    "neonatal rat ventricular cardiomyocytes",
-    "MG-63",
-    "Neonatal mouse cardiomyocytes (NMVCMs)",
-    "human hepatic stellate cell line",
-    "HEK-293",
-    "aHSC",
-    "MFCs",
-    "fibroblasts",
-    "HNDF",
-    "cardiomyocyte/MSCs",
-    "ADSCs",
-    "HCASMCs",
-    "cardiomyocyte",
-    "hCPCs",
-    "Human CM /adult human fibroblasts ",
-    "primary rat hepatocyte",
-    "human cardiac progenitor cells",
-    "SMC",
-    "Human MSCs",
-    "ACPCs",
-    "Huh7/HepaRG",
-    "Human umbilical vein endothelial cells",
-    "ATDC5",
-    "hESC-derived HLCs",
-    "NIH 3T3",
-    "n neonatal mouse ventricular cardiomyocytes",
-    "CFs/CMs/HUVECs",
-    "MRC-5",
-    "VIC",
-    "eHep",
-    "hUVECs/NIH3T3",
-    "MC3T3",
-    "HLC",
-    "hepatoma",
-    "FB",
-    "A549 GFP+",
-    "HPAAF",
-    "PMHs",
-    "HUVSMCs",
-    "Human CPCs",
-    "Fibroblasts/THP-1",
-    "rat ventricular cardiomyocytes",
-    "iCMs/iCFs/iECs",
-    "HUVEC/HHSC",
-    "Human CPCs / MSCs",
-    "HepaRG/LX-2 ",
-    "iCMs/iCFs/iECs/iCMFs",
-    "LX-2 ",
-    "SMMC-7721",
-    "Hepatoblast- single cell/iESC/iMSC",
-    "iCMFs",
-    "Hepatoblast- spheroid/iESC/iMSC",
-    "hBM-MSCs",
-    "BMSCs",
-    "HUVECs and HHSCs",
-    "Intrahepatic cholangiocarcinoma (ICC)",
-    "VICs",
-    "CMs/CFs",
-    "human neonatal dermal fi broblasts",
-    "10T1/2 fibroblast-laden cells",
-    "human cardiac fibroblasts",
-    "neonatal rat ventricular CMs",
-    "HUVECs and hiPSC-CS",
-    "iPSC-derived CM",
-    "Human Umbilical Vein Endothelial Cells + iPSC-derived CM",
-    "rabbit bone marrow mesenchymal stem cells",
-    "Neonatal rat cardiomyocytes",
-    "NIH 3T3 mouse fibroblasts",
-    "Human CPCs & MSCs",
-    "NSCLC PDX/CAFs",
-    "hiPSCs-derived HLCs",
-    "U87",
-    "75% hepatoblast cells, 20% iEC and 5% iMSC",
-    "SMCs",
-    "A549/95-D cells",
-    "NCI-H441",
-    "pancreatic cancer cell",
-    "prostate cancer stem cell",
-    "human primary parathyroid cells ",
-    "primary human hepatocytes",
-    "CPCs / MSCs",
-    "CPCs ",
-    "cardiac fibroblasts",
-    "iPSCs-derived cardiomyocytes",
-    "cardiomyocytes/ fibroblasts",
-    "hiPSC-CM",
-    "iPSCs/HUVECs",
-    "iPSC-CMs",
-    "iPSCs",
-    "H1395",
-    "PC9",
-    "H1650",
-    "HULEC-5a",
-    "NCI-H1703"
-]
+    'NoCellCultured',
+    'HUVECs',
+    'HepG2',
+    'bMSCs',
+    'chondrocytes',
+    'NIH3T3',
+    'HepaRG',
+    'hiPSC-CMs /ATCCs',
+    'Myoblast cells',
+    'H9c2',
+    'HEK',
+    'CPCs',
+    'MESCs',
+    'hiPSCs',
+    'L929',
+    'Primary mouse hepatocyte',
+    'NSCLC PDX',
+    'human primary kidney cells',
+    'Cardiac progenitor cells',
+    '10T1/2',
+    'HAVIC',
+    'RAMECs',
+    'hESCs',
+    'human nasoseptal chondrocytes',
+    'PDX',
+    'PTECs',
+    'HPFs',
+    'MG-63',
+    'dermal fibroblasts',
+    'HASSMC',
+    'Saos-2',
+    'MC3T3-E1',
+    'hASCs',
+    'SU3',
+    'ESCs',
+    'RPTEC/TERT1',
+    'hiPSC-CMs and HS-27A',
+    'HADMSCs',
+    'HACs',
+    'MRC5',
+    'HeLa',
+    'RHECs',
+    'cardiomyocytes',
+    'human adipogenic mesenchymal stem cells',
+    'MSCs',
+    'HUVECs/MSCs',
+    'U87-MG',
+    'RSC96/HUVECs',
+    'HEPG2/ECs',
+    'A549',
+    'Human non-small cell lung cancer line Calu-3 (Calu-3)',
+    'hTMSCs',
+    'mouse cardiac cells',
+    'Human cardiac fibroblasts',
+    'Schwann cells',
+    'hTERT-MSC',
+    'bMSCs/hACs',
+    'basil plant cell',
+    'EPCs',
+    'HepaRG / LX-2',
+    'Caki-1',
+    'human induced pluripotent stem cell-derived cardiomyocytes',
+    'human epithelial lung carcinoma cells',
+    'HCF',
+    'Huh7',
+    '3T3 Fibroblasts',
+    'EA.hy 926 cells',
+    'human coronary artery endothelial cells',
+    'NIH3T3/ HUVECs',
+    'neonatal rat ventricular CFs',
+    'HepG2/C3A',
+    'IMR-90',
+    'rMSC',
+    'hiPSC-CM / fibroblasts',
+    'HUVECs/H9C2',
+    'fibroblasts',
+    'Endothelial cells',
+    'neonatal rat ventricular cardiomyocytes',
+    'Neonatal mouse cardiomyocytes (NMVCMs)',
+    'murine macrophage-like cell line',
+    '293FT',
+    'sADSC',
+    'human hepatic stellate cell line',
+    'aHSC',
+    'NRCMs',
+    'HBE',
+    'HEK-293',
+    'Wnt3a-293FT',
+    'Human aortic VIC',
+    'HEK 293FT',
+    'HNDF',
+    'Huh7/HepaRG',
+    'hiHeps',
+    'hCPCs',
+    'pancreatic islets cells',
+    'ICC',
+    'Wnt reporter-293FT',
+    'Human CPCs',
+    'human cardiac progenitor cells',
+    'SMC',
+    'hepatoma',
+    'cardiomyocyte/MSCs',
+    'FB',
+    'eHep',
+    'PMHs',
+    'Human CM /adult human fibroblasts',
+    'HUVSMCs',
+    'ACPCs',
+    'MC3T3',
+    'hNCs',
+    'primary rat hepatocyte',
+    'Human umbilical vein endothelial cells',
+    'HPAAF',
+    'VIC',
+    'HLC',
+    'ADSCs',
+    'MFCs',
+    'A549 GFP+',
+    'ATDC5',
+    'CFs/CMs/HUVECs',
+    'n neonatal mouse ventricular cardiomyocytes',
+    'Human MSCs',
+    'Human CPCs / MSCs',
+    'hESC-derived HLCs',
+    'HCASMCs',
+    'iCMs/iCFs/iECs/iCMFs',
+    'VICs',
+    'iPSCs',
+    'primary human hepatocytes',
+    'SMCs',
+    'iCMs/iCFs/iECs',
+    'PC9',
+    'H1395',
+    'H1650',
+    'HUVECs and hiPSC-CS',
+    'human neonatal dermal fi broblasts',
+    '10T1/2 fibroblast-laden cells',
+    'iCMFs',
+    'neonatal rat ventricular CMs',
+    'U87',
+    'NSCLC PDX/CAFs',
+    'iPSC-derived CM',
+    'NIH 3T3 mouse fibroblasts',
+    'Neonatal rat cardiomyocytes',
+    'hiPSCs-derived HLCs',
+    'iPSCs/HUVECs',
+    'HULEC-5a',
+    'HUVECs and HHSCs',
+    'NCI-H1703',
+    '75% hepatoblast cells, 20% iEC and 5% iMSC',
+    'Intrahepatic cholangiocarcinoma (ICC)',
+    'NCI-H441',
+    'Human Umbilical Vein Endothelial Cells + iPSC-derived CM',
+    'rabbit bone marrow mesenchymal stem cells',
+    'CMs/CFs',
+    'cardiac fibroblasts',
+    'iPSCs-derived cardiomyocytes',
+    'cardiomyocytes/ fibroblasts',
+    'hiPSC-CM',
+    'iPSC-CMs',
+    'hBM-MSCs',
+    'Hepatoblast- single cell/iESC/iMSC',
+    'Hepatoblast- spheroid/iESC/iMSC',
+    'human primary parathyroid cells',
+    'prostate cancer stem cell',
+    'HL-1',
+    'CPCs / MSCs',
+    'pancreatic cancer cell',
+    'A549/95-D cells',
+    'SMMC-7721',
+    'rat ventricular cardiomyocytes',
+    'A549 lung adenocarcinoma cells',
+    'HMEC-1/ fibroblast/ THP-1',
+    'HMEC-1',
+    'HAVSMC',
+    'HUVEC/HHSC',
+    'LX-2',
+    'Fibroblasts/THP-1',
+    'HLF',
+    'HUVECs / LX-2',
+    'CCAO',
+    'Hepa RG/ HUVEC',
+    'BEAS2B',
+    'PDGFRβ+',
+    'hMSCs',
+    'CiPSC-HOs',
+    'tubular epithelial',
+]
+
+CELL_LINE_COUNTS = {
+    'NoCellCultured': 1573,
+    'HUVECs': 75,
+    'HepG2': 71,
+    'bMSCs': 65,
+    'chondrocytes': 39,
+    'NIH3T3': 35,
+    'HepaRG': 28,
+    'hiPSC-CMs /ATCCs': 26,
+    'Myoblast cells': 26,
+    'H9c2': 25,
+    'HEK': 24,
+    'CPCs': 21,
+    'MESCs': 20,
+    'hiPSCs': 17,
+    'L929': 15,
+    'Primary mouse hepatocyte': 14,
+    'NSCLC PDX': 13,
+    'human primary kidney cells': 13,
+    'Cardiac progenitor cells': 13,
+    '10T1/2': 13,
+    'HAVIC': 12,
+    'RAMECs': 12,
+    'hESCs': 11,
+    'human nasoseptal chondrocytes': 10,
+    'PDX': 10,
+    'PTECs': 10,
+    'HPFs': 10,
+    'MG-63': 9,
+    'dermal fibroblasts': 9,
+    'HASSMC': 9,
+    'Saos-2': 8,
+    'MC3T3-E1': 8,
+    'hASCs': 8,
+    'SU3': 8,
+    'ESCs': 8,
+    'RPTEC/TERT1': 8,
+    'hiPSC-CMs and HS-27A': 8,
+    'HADMSCs': 7,
+    'HACs': 7,
+    'MRC5': 7,
+    'HeLa': 7,
+    'RHECs': 6,
+    'cardiomyocytes': 6,
+    'human adipogenic mesenchymal stem cells': 6,
+    'MSCs': 6,
+    'HUVECs/MSCs': 6,
+    'U87-MG': 6,
+    'RSC96/HUVECs': 6,
+    'HEPG2/ECs': 6,
+    'A549': 6,
+    'Human non-small cell lung cancer line Calu-3 (Calu-3)': 5,
+    'hTMSCs': 5,
+    'mouse cardiac cells': 5,
+    'Human cardiac fibroblasts': 5,
+    'Schwann cells': 5,
+    'hTERT-MSC': 4,
+    'bMSCs/hACs': 4,
+    'basil plant cell': 4,
+    'EPCs': 4,
+    'HepaRG / LX-2': 4,
+    'Caki-1': 4,
+    'human induced pluripotent stem cell-derived cardiomyocytes': 4,
+    'human epithelial lung carcinoma cells': 4,
+    'HCF': 4,
+    'Huh7': 4,
+    '3T3 Fibroblasts': 4,
+    'EA.hy 926 cells': 4,
+    'human coronary artery endothelial cells': 4,
+    'NIH3T3/ HUVECs': 4,
+    'neonatal rat ventricular CFs': 4,
+    'HepG2/C3A': 4,
+    'IMR-90': 4,
+    'rMSC': 4,
+    'hiPSC-CM / fibroblasts': 3,
+    'HUVECs/H9C2': 3,
+    'fibroblasts': 3,
+    'Endothelial cells': 3,
+    'neonatal rat ventricular cardiomyocytes': 3,
+    'Neonatal mouse cardiomyocytes (NMVCMs)': 3,
+    'murine macrophage-like cell line': 3,
+    '293FT': 3,
+    'sADSC': 3,
+    'human hepatic stellate cell line': 3,
+    'aHSC': 3,
+    'NRCMs': 3,
+    'HBE': 3,
+    'HEK-293': 3,
+    'Wnt3a-293FT': 3,
+    'Human aortic VIC': 3,
+    'HEK 293FT': 3,
+    'HNDF': 2,
+    'Huh7/HepaRG': 2,
+    'hiHeps': 2,
+    'hCPCs': 2,
+    'pancreatic islets cells': 2,
+    'ICC': 2,
+    'Wnt reporter-293FT': 2,
+    'Human CPCs': 2,
+    'human cardiac progenitor cells': 2,
+    'SMC': 2,
+    'hepatoma': 2,
+    'cardiomyocyte/MSCs': 2,
+    'FB': 2,
+    'eHep': 2,
+    'PMHs': 2,
+    'Human CM /adult human fibroblasts': 2,
+    'HUVSMCs': 2,
+    'ACPCs': 2,
+    'MC3T3': 2,
+    'hNCs': 2,
+    'primary rat hepatocyte': 2,
+    'Human umbilical vein endothelial cells': 2,
+    'HPAAF': 2,
+    'VIC': 2,
+    'HLC': 2,
+    'ADSCs': 2,
+    'MFCs': 2,
+    'A549 GFP+': 2,
+    'ATDC5': 2,
+    'CFs/CMs/HUVECs': 2,
+    'n neonatal mouse ventricular cardiomyocytes': 2,
+    'Human MSCs': 2,
+    'Human CPCs / MSCs': 2,
+    'hESC-derived HLCs': 2,
+    'HCASMCs': 2,
+    'iCMs/iCFs/iECs/iCMFs': 1,
+    'VICs': 1,
+    'iPSCs': 1,
+    'primary human hepatocytes': 1,
+    'SMCs': 1,
+    'iCMs/iCFs/iECs': 1,
+    'PC9': 1,
+    'H1395': 1,
+    'H1650': 1,
+    'HUVECs and hiPSC-CS': 1,
+    'human neonatal dermal fi broblasts': 1,
+    '10T1/2 fibroblast-laden cells': 1,
+    'iCMFs': 1,
+    'neonatal rat ventricular CMs': 1,
+    'U87': 1,
+    'NSCLC PDX/CAFs': 1,
+    'iPSC-derived CM': 1,
+    'NIH 3T3 mouse fibroblasts': 1,
+    'Neonatal rat cardiomyocytes': 1,
+    'hiPSCs-derived HLCs': 1,
+    'iPSCs/HUVECs': 1,
+    'HULEC-5a': 1,
+    'HUVECs and HHSCs': 1,
+    'NCI-H1703': 1,
+    '75% hepatoblast cells, 20% iEC and 5% iMSC': 1,
+    'Intrahepatic cholangiocarcinoma (ICC)': 1,
+    'NCI-H441': 1,
+    'Human Umbilical Vein Endothelial Cells + iPSC-derived CM': 1,
+    'rabbit bone marrow mesenchymal stem cells': 1,
+    'CMs/CFs': 1,
+    'cardiac fibroblasts': 1,
+    'iPSCs-derived cardiomyocytes': 1,
+    'cardiomyocytes/ fibroblasts': 1,
+    'hiPSC-CM': 1,
+    'iPSC-CMs': 1,
+    'hBM-MSCs': 1,
+    'Hepatoblast- single cell/iESC/iMSC': 1,
+    'Hepatoblast- spheroid/iESC/iMSC': 1,
+    'human primary parathyroid cells': 1,
+    'prostate cancer stem cell': 1,
+    'HL-1': 1,
+    'CPCs / MSCs': 1,
+    'pancreatic cancer cell': 1,
+    'A549/95-D cells': 1,
+    'SMMC-7721': 1,
+    'rat ventricular cardiomyocytes': 1,
+    'A549 lung adenocarcinoma cells': 1,
+    'HMEC-1/ fibroblast/ THP-1': 1,
+    'HMEC-1': 1,
+    'HAVSMC': 1,
+    'HUVEC/HHSC': 1,
+    'LX-2': 1,
+    'Fibroblasts/THP-1': 1,
+    'HLF': 1,
+    'HUVECs / LX-2': 1,
+    'CCAO': 1,
+    'Hepa RG/ HUVEC': 1,
+    'BEAS2B': 1,
+    'PDGFRβ+': 1,
+    'hMSCs': 1,
+    'CiPSC-HOs': 1,
+    'tubular epithelial': 1,
+}
+
+# Density in 10^6 cells/mL, among records using that line.
+CELL_DENSITY_RANGES = {
+    "NoCellCultured": {
+        "min": 1.0,
+        "max": 3.0,
+        "median": 1.0
+    },
+    "HUVECs": {
+        "min": 0.035,
+        "max": 20.0,
+        "median": 5.0
+    },
+    "HepG2": {
+        "min": 1.0,
+        "max": 20.0,
+        "median": 1.5
+    },
+    "bMSCs": {
+        "min": 0.25,
+        "max": 20.0,
+        "median": 10.0
+    },
+    "chondrocytes": {
+        "min": 0.1,
+        "max": 6.0,
+        "median": 6.0
+    },
+    "NIH3T3": {
+        "min": 0.1,
+        "max": 5.0,
+        "median": 2.5
+    },
+    "HepaRG": {
+        "min": 1.0,
+        "max": 20.0,
+        "median": 4.0
+    },
+    "hiPSC-CMs /ATCCs": {
+        "min": 2.0,
+        "max": 2.0,
+        "median": 2.0
+    },
+    "Myoblast cells": {
+        "min": 0.366,
+        "max": 10.0,
+        "median": 1.0
+    },
+    "H9c2": {
+        "min": 0.04,
+        "max": 1.0,
+        "median": 1.0
+    },
+    "HEK": {
+        "min": 3.0,
+        "max": 3.0,
+        "median": 3.0
+    },
+    "CPCs": {
+        "min": 2.0,
+        "max": 8.0,
+        "median": 2.0
+    },
+    "MESCs": {
+        "min": 0.5,
+        "max": 2.0,
+        "median": 1.0
+    },
+    "hiPSCs": {
+        "min": 1.0,
+        "max": 15.0,
+        "median": 15.0
+    },
+    "L929": {
+        "min": 0.1,
+        "max": 10.0,
+        "median": 1.0
+    },
+    "Primary mouse hepatocyte": {
+        "min": 0.128,
+        "max": 1.0,
+        "median": 0.128
+    },
+    "NSCLC PDX": {
+        "min": 10.0,
+        "max": 10.0,
+        "median": 10.0
+    },
+    "human primary kidney cells": {
+        "min": 10.0,
+        "max": 10.0,
+        "median": 10.0
+    },
+    "Cardiac progenitor cells": {
+        "min": 5.0,
+        "max": 5.0,
+        "median": 5.0
+    },
+    "10T1/2": {
+        "min": 0.5,
+        "max": 2.0,
+        "median": 0.5
+    },
+    "HAVIC": {
+        "min": 2.5,
+        "max": 10.0,
+        "median": 3.75
+    },
+    "RAMECs": {
+        "min": 1.0,
+        "max": 1.0,
+        "median": 1.0
+    },
+    "hESCs": {
+        "min": 1.0,
+        "max": 1.0,
+        "median": 1.0
+    },
+    "human nasoseptal chondrocytes": {
+        "min": 4.0,
+        "max": 4.0,
+        "median": 4.0
+    },
+    "PDX": {
+        "min": 10.0,
+        "max": 10.0,
+        "median": 10.0
+    },
+    "PTECs": {
+        "min": 15.0,
+        "max": 15.0,
+        "median": 15.0
+    },
+    "HPFs": {
+        "min": 2.0,
+        "max": 2.0,
+        "median": 2.0
+    },
+    "MG-63": {
+        "min": 0.005,
+        "max": 5.0,
+        "median": 0.005
+    },
+    "dermal fibroblasts": {
+        "min": 10.0,
+        "max": 10.0,
+        "median": 10.0
+    },
+    "HASSMC": {
+        "min": 2.5,
+        "max": 2.5,
+        "median": 2.5
+    },
+    "Saos-2": {
+        "min": 5.0,
+        "max": 5.0,
+        "median": 5.0
+    },
+    "MC3T3-E1": {
+        "min": 30.0,
+        "max": 30.0,
+        "median": 30.0
+    },
+    "hASCs": {
+        "min": 1.0,
+        "max": 2.0,
+        "median": 1.3
+    },
+    "SU3": {
+        "min": 5.0,
+        "max": 5.0,
+        "median": 5.0
+    },
+    "ESCs": {
+        "min": 6.0,
+        "max": 10.0,
+        "median": 8.0
+    },
+    "RPTEC/TERT1": {
+        "min": 10.0,
+        "max": 10.0,
+        "median": 10.0
+    },
+    "hiPSC-CMs and HS-27A": {
+        "min": 20.0,
+        "max": 20.0,
+        "median": 20.0
+    },
+    "HADMSCs": {
+        "min": 2.0,
+        "max": 2.5,
+        "median": 2.5
+    },
+    "HACs": {
+        "min": 5.0,
+        "max": 15.0,
+        "median": 5.0
+    },
+    "MRC5": {
+        "min": 0.04,
+        "max": 5.0,
+        "median": 3.0
+    },
+    "HeLa": {
+        "min": 1.0,
+        "max": 150.0,
+        "median": 1.0
+    },
+    "RHECs": {
+        "min": 0.5,
+        "max": 0.5,
+        "median": 0.5
+    },
+    "cardiomyocytes": {
+        "min": 0.2,
+        "max": 10.0,
+        "median": 6.0
+    },
+    "human adipogenic mesenchymal stem cells": {
+        "min": 0.75,
+        "max": 2.25,
+        "median": 1.125
+    },
+    "MSCs": {
+        "min": 0.2,
+        "max": 5.0,
+        "median": 1.0
+    },
+    "HUVECs/MSCs": {
+        "min": 6.0,
+        "max": 6.0,
+        "median": 6.0
+    },
+    "U87-MG": {
+        "min": 5.25,
+        "max": 5.25,
+        "median": 5.25
+    },
+    "RSC96/HUVECs": {
+        "min": 1.0,
+        "max": 1.0,
+        "median": 1.0
+    },
+    "HEPG2/ECs": {
+        "min": 6.0,
+        "max": 6.0,
+        "median": 6.0
+    },
+    "A549": {
+        "min": 1.0,
+        "max": 5.0,
+        "median": 2.5
+    },
+    "Human non-small cell lung cancer line Calu-3 (Calu-3)": {
+        "min": 10.0,
+        "max": 10.0,
+        "median": 10.0
+    },
+    "hTMSCs": {
+        "min": 3.5,
+        "max": 3.5,
+        "median": 3.5
+    },
+    "mouse cardiac cells": {
+        "min": 0.32,
+        "max": 0.32,
+        "median": 0.32
+    },
+    "Human cardiac fibroblasts": {
+        "min": 1.0,
+        "max": 1.0,
+        "median": 1.0
+    },
+    "Schwann cells": {
+        "min": 1.0,
+        "max": 2.0,
+        "median": 1.0
+    },
+    "hTERT-MSC": {
+        "min": 10.0,
+        "max": 10.0,
+        "median": 10.0
+    },
+    "bMSCs/hACs": {
+        "min": 10.0,
+        "max": 10.0,
+        "median": 10.0
+    },
+    "EPCs": {
+        "min": 10.0,
+        "max": 10.0,
+        "median": 10.0
+    },
+    "HepaRG / LX-2": {
+        "min": 1.5,
+        "max": 3.0,
+        "median": 3.0
+    },
+    "Caki-1": {
+        "min": 10.0,
+        "max": 10.0,
+        "median": 10.0
+    },
+    "human induced pluripotent stem cell-derived cardiomyocytes": {
+        "min": 20.0,
+        "max": 20.0,
+        "median": 20.0
+    },
+    "human epithelial lung carcinoma cells": {
+        "min": 7.0,
+        "max": 7.0,
+        "median": 7.0
+    },
+    "HCF": {
+        "min": 2.0,
+        "max": 2.0,
+        "median": 2.0
+    },
+    "Huh7": {
+        "min": 0.5,
+        "max": 0.5,
+        "median": 0.5
+    },
+    "3T3 Fibroblasts": {
+        "min": 5.0,
+        "max": 5.0,
+        "median": 5.0
+    },
+    "EA.hy 926 cells": {
+        "min": 5.0,
+        "max": 5.0,
+        "median": 5.0
+    },
+    "human coronary artery endothelial cells": {
+        "min": 0.8,
+        "max": 1.0,
+        "median": 0.8
+    },
+    "NIH3T3/ HUVECs": {
+        "min": 1.0,
+        "max": 1.0,
+        "median": 1.0
+    },
+    "neonatal rat ventricular CFs": {
+        "min": 5.0,
+        "max": 5.0,
+        "median": 5.0
+    },
+    "HepG2/C3A": {
+        "min": 5.0,
+        "max": 5.0,
+        "median": 5.0
+    },
+    "IMR-90": {
+        "min": 3.5,
+        "max": 3.5,
+        "median": 3.5
+    },
+    "rMSC": {
+        "min": 3.5,
+        "max": 3.5,
+        "median": 3.5
+    },
+    "hiPSC-CM / fibroblasts": {
+        "min": 1.0,
+        "max": 1.0,
+        "median": 1.0
+    },
+    "HUVECs/H9C2": {
+        "min": 4.0,
+        "max": 4.0,
+        "median": 4.0
+    },
+    "fibroblasts": {
+        "min": 1.0,
+        "max": 10.0,
+        "median": 3.0
+    },
+    "neonatal rat ventricular cardiomyocytes": {
+        "min": 1.0,
+        "max": 200.0,
+        "median": 1.0
+    },
+    "Neonatal mouse cardiomyocytes (NMVCMs)": {
+        "min": 10.0,
+        "max": 10.0,
+        "median": 10.0
+    },
+    "murine macrophage-like cell line": {
+        "min": 1.0,
+        "max": 1.0,
+        "median": 1.0
+    },
+    "293FT": {
+        "min": 2.0,
+        "max": 2.0,
+        "median": 2.0
+    },
+    "sADSC": {
+        "min": 1.0,
+        "max": 3.0,
+        "median": 2.0
+    },
+    "human hepatic stellate cell line": {
+        "min": 3.0,
+        "max": 5.0,
+        "median": 4.0
+    },
+    "aHSC": {
+        "min": 3.0,
+        "max": 5.0,
+        "median": 4.0
+    },
+    "NRCMs": {
+        "min": 20.0,
+        "max": 20.0,
+        "median": 20.0
+    },
+    "HBE": {
+        "min": 0.13,
+        "max": 0.13,
+        "median": 0.13
+    },
+    "HEK-293": {
+        "min": 0.05,
+        "max": 2.0,
+        "median": 2.0
+    },
+    "Wnt3a-293FT": {
+        "min": 2.0,
+        "max": 2.0,
+        "median": 2.0
+    },
+    "Human aortic VIC": {
+        "min": 5.0,
+        "max": 5.0,
+        "median": 5.0
+    },
+    "HEK 293FT": {
+        "min": 2.0,
+        "max": 2.0,
+        "median": 2.0
+    },
+    "HNDF": {
+        "min": 2.0,
+        "max": 2.0,
+        "median": 2.0
+    },
+    "Huh7/HepaRG": {
+        "min": 1.5,
+        "max": 1.5,
+        "median": 1.5
+    },
+    "hiHeps": {
+        "min": 10.0,
+        "max": 50.0,
+        "median": 30.0
+    },
+    "hCPCs": {
+        "min": 3.0,
+        "max": 3.0,
+        "median": 3.0
+    },
+    "pancreatic islets cells": {
+        "min": 0.045,
+        "max": 0.045,
+        "median": 0.045
+    },
+    "ICC": {
+        "min": 3.0,
+        "max": 6.0,
+        "median": 4.5
+    },
+    "Wnt reporter-293FT": {
+        "min": 2.0,
+        "max": 2.0,
+        "median": 2.0
+    },
+    "Human CPCs": {
+        "min": 5e-06,
+        "max": 5.0,
+        "median": 2.5000025
+    },
+    "human cardiac progenitor cells": {
+        "min": 3.0,
+        "max": 3.0,
+        "median": 3.0
+    },
+    "SMC": {
+        "min": 2.0,
+        "max": 10.0,
+        "median": 6.0
+    },
+    "hepatoma": {
+        "min": 1.0,
+        "max": 1.0,
+        "median": 1.0
+    },
+    "cardiomyocyte/MSCs": {
+        "min": 10.0,
+        "max": 10.0,
+        "median": 10.0
+    },
+    "FB": {
+        "min": 1.0,
+        "max": 1.0,
+        "median": 1.0
+    },
+    "eHep": {
+        "min": 5.0,
+        "max": 5.0,
+        "median": 5.0
+    },
+    "PMHs": {
+        "min": 10.0,
+        "max": 10.0,
+        "median": 10.0
+    },
+    "Human CM /adult human fibroblasts": {
+        "min": 0.01,
+        "max": 0.2,
+        "median": 0.10500000000000001
+    },
+    "HUVSMCs": {
+        "min": 10.0,
+        "max": 10.0,
+        "median": 10.0
+    },
+    "ACPCs": {
+        "min": 20.0,
+        "max": 20.0,
+        "median": 20.0
+    },
+    "MC3T3": {
+        "min": 0.25,
+        "max": 0.25,
+        "median": 0.25
+    },
+    "hNCs": {
+        "min": 1.0,
+        "max": 15.0,
+        "median": 8.0
+    },
+    "primary rat hepatocyte": {
+        "min": 1.0,
+        "max": 5.0,
+        "median": 3.0
+    },
+    "Human umbilical vein endothelial cells": {
+        "min": 1.25,
+        "max": 10.0,
+        "median": 5.625
+    },
+    "HPAAF": {
+        "min": 0.4,
+        "max": 4.0,
+        "median": 2.2
+    },
+    "VIC": {
+        "min": 2.0,
+        "max": 10.0,
+        "median": 6.0
+    },
+    "HLC": {
+        "min": 2.0,
+        "max": 2.0,
+        "median": 2.0
+    },
+    "ADSCs": {
+        "min": 2.0,
+        "max": 2.0,
+        "median": 2.0
+    },
+    "MFCs": {
+        "min": 5.0,
+        "max": 5.0,
+        "median": 5.0
+    },
+    "A549 GFP+": {
+        "min": 1.5,
+        "max": 1.5,
+        "median": 1.5
+    },
+    "ATDC5": {
+        "min": 5.68,
+        "max": 5.68,
+        "median": 5.68
+    },
+    "CFs/CMs/HUVECs": {
+        "min": 30.0,
+        "max": 30.0,
+        "median": 30.0
+    },
+    "n neonatal mouse ventricular cardiomyocytes": {
+        "min": 10.0,
+        "max": 10.0,
+        "median": 10.0
+    },
+    "Human MSCs": {
+        "min": 5e-06,
+        "max": 5.0,
+        "median": 2.5000025
+    },
+    "Human CPCs / MSCs": {
+        "min": 5.0,
+        "max": 5.0,
+        "median": 5.0
+    },
+    "hESC-derived HLCs": {
+        "min": 10.0,
+        "max": 10.0,
+        "median": 10.0
+    },
+    "HCASMCs": {
+        "min": 10.0,
+        "max": 10.0,
+        "median": 10.0
+    },
+    "iCMs/iCFs/iECs/iCMFs": {
+        "min": 3.87,
+        "max": 3.87,
+        "median": 3.87
+    },
+    "VICs": {
+        "min": 10.0,
+        "max": 10.0,
+        "median": 10.0
+    },
+    "iPSCs": {
+        "min": 8.0,
+        "max": 8.0,
+        "median": 8.0
+    },
+    "primary human hepatocytes": {
+        "min": 5.0,
+        "max": 5.0,
+        "median": 5.0
+    },
+    "SMCs": {
+        "min": 10.0,
+        "max": 10.0,
+        "median": 10.0
+    },
+    "iCMs/iCFs/iECs": {
+        "min": 9.6,
+        "max": 9.6,
+        "median": 9.6
+    },
+    "PC9": {
+        "min": 1.0,
+        "max": 1.0,
+        "median": 1.0
+    },
+    "H1395": {
+        "min": 1.0,
+        "max": 1.0,
+        "median": 1.0
+    },
+    "H1650": {
+        "min": 1.0,
+        "max": 1.0,
+        "median": 1.0
+    },
+    "HUVECs and hiPSC-CS": {
+        "min": 0.01,
+        "max": 0.01,
+        "median": 0.01
+    },
+    "human neonatal dermal fi broblasts": {
+        "min": 2.0,
+        "max": 2.0,
+        "median": 2.0
+    },
+    "10T1/2 fibroblast-laden cells": {
+        "min": 2.0,
+        "max": 2.0,
+        "median": 2.0
+    },
+    "iCMFs": {
+        "min": 10.0,
+        "max": 10.0,
+        "median": 10.0
+    },
+    "neonatal rat ventricular CMs": {
+        "min": 5.0,
+        "max": 5.0,
+        "median": 5.0
+    },
+    "U87": {
+        "min": 5.0,
+        "max": 5.0,
+        "median": 5.0
+    },
+    "NSCLC PDX/CAFs": {
+        "min": 5.0,
+        "max": 5.0,
+        "median": 5.0
+    },
+    "iPSC-derived CM": {
+        "min": 8.0,
+        "max": 8.0,
+        "median": 8.0
+    },
+    "NIH 3T3 mouse fibroblasts": {
+        "min": 5.0,
+        "max": 5.0,
+        "median": 5.0
+    },
+    "Neonatal rat cardiomyocytes": {
+        "min": 0.2,
+        "max": 0.2,
+        "median": 0.2
+    },
+    "hiPSCs-derived HLCs": {
+        "min": 10.0,
+        "max": 10.0,
+        "median": 10.0
+    },
+    "iPSCs/HUVECs": {
+        "min": 10.0,
+        "max": 10.0,
+        "median": 10.0
+    },
+    "HULEC-5a": {
+        "min": 6.0,
+        "max": 6.0,
+        "median": 6.0
+    },
+    "HUVECs and HHSCs": {
+        "min": 1.8,
+        "max": 1.8,
+        "median": 1.8
+    },
+    "NCI-H1703": {
+        "min": 6.0,
+        "max": 6.0,
+        "median": 6.0
+    },
+    "75% hepatoblast cells, 20% iEC and 5% iMSC": {
+        "min": 10.0,
+        "max": 10.0,
+        "median": 10.0
+    },
+    "Intrahepatic cholangiocarcinoma (ICC)": {
+        "min": 1.0,
+        "max": 1.0,
+        "median": 1.0
+    },
+    "NCI-H441": {
+        "min": 6.0,
+        "max": 6.0,
+        "median": 6.0
+    },
+    "Human Umbilical Vein Endothelial Cells + iPSC-derived CM": {
+        "min": 6.0,
+        "max": 6.0,
+        "median": 6.0
+    },
+    "rabbit bone marrow mesenchymal stem cells": {
+        "min": 0.1,
+        "max": 0.1,
+        "median": 0.1
+    },
+    "CMs/CFs": {
+        "min": 30.0,
+        "max": 30.0,
+        "median": 30.0
+    },
+    "cardiac fibroblasts": {
+        "min": 5.0,
+        "max": 5.0,
+        "median": 5.0
+    },
+    "iPSCs-derived cardiomyocytes": {
+        "min": 200.0,
+        "max": 200.0,
+        "median": 200.0
+    },
+    "cardiomyocytes/ fibroblasts": {
+        "min": 0.2,
+        "max": 0.2,
+        "median": 0.2
+    },
+    "hiPSC-CM": {
+        "min": 0.2,
+        "max": 0.2,
+        "median": 0.2
+    },
+    "iPSC-CMs": {
+        "min": 4.0,
+        "max": 4.0,
+        "median": 4.0
+    },
+    "hBM-MSCs": {
+        "min": 0.25,
+        "max": 0.25,
+        "median": 0.25
+    },
+    "Hepatoblast- single cell/iESC/iMSC": {
+        "min": 5.6,
+        "max": 5.6,
+        "median": 5.6
+    },
+    "Hepatoblast- spheroid/iESC/iMSC": {
+        "min": 5.6,
+        "max": 5.6,
+        "median": 5.6
+    },
+    "human primary parathyroid cells": {
+        "min": 0.06,
+        "max": 0.06,
+        "median": 0.06
+    },
+    "prostate cancer stem cell": {
+        "min": 2.0,
+        "max": 2.0,
+        "median": 2.0
+    },
+    "HL-1": {
+        "min": 8.0,
+        "max": 8.0,
+        "median": 8.0
+    },
+    "CPCs / MSCs": {
+        "min": 5.0,
+        "max": 5.0,
+        "median": 5.0
+    },
+    "pancreatic cancer cell": {
+        "min": 2.0,
+        "max": 2.0,
+        "median": 2.0
+    },
+    "A549/95-D cells": {
+        "min": 1.0,
+        "max": 1.0,
+        "median": 1.0
+    },
+    "SMMC-7721": {
+        "min": 0.5,
+        "max": 0.5,
+        "median": 0.5
+    },
+    "rat ventricular cardiomyocytes": {
+        "min": 10.0,
+        "max": 10.0,
+        "median": 10.0
+    },
+    "A549 lung adenocarcinoma cells": {
+        "min": 10.0,
+        "max": 10.0,
+        "median": 10.0
+    },
+    "HMEC-1/ fibroblast/ THP-1": {
+        "min": 35.0,
+        "max": 35.0,
+        "median": 35.0
+    },
+    "HMEC-1": {
+        "min": 25.0,
+        "max": 25.0,
+        "median": 25.0
+    },
+    "HAVSMC": {
+        "min": 1.5,
+        "max": 1.5,
+        "median": 1.5
+    },
+    "HUVEC/HHSC": {
+        "min": 0.9,
+        "max": 0.9,
+        "median": 0.9
+    },
+    "LX-2": {
+        "min": 0.5,
+        "max": 0.5,
+        "median": 0.5
+    },
+    "Fibroblasts/THP-1": {
+        "min": 14.25,
+        "max": 14.25,
+        "median": 14.25
+    },
+    "HLF": {
+        "min": 5.0,
+        "max": 5.0,
+        "median": 5.0
+    },
+    "HUVECs / LX-2": {
+        "min": 10.0,
+        "max": 10.0,
+        "median": 10.0
+    },
+    "CCAO": {
+        "min": 0.02,
+        "max": 0.02,
+        "median": 0.02
+    },
+    "Hepa RG/ HUVEC": {
+        "min": 20.0,
+        "max": 20.0,
+        "median": 20.0
+    },
+    "BEAS2B": {
+        "min": 2.0,
+        "max": 2.0,
+        "median": 2.0
+    },
+    "PDGFR\u03b2+": {
+        "min": 0.25,
+        "max": 0.25,
+        "median": 0.25
+    },
+    "hMSCs": {
+        "min": 2.0,
+        "max": 2.0,
+        "median": 2.0
+    },
+    "CiPSC-HOs": {
+        "min": 2.5,
+        "max": 2.5,
+        "median": 2.5
+    },
+    "tubular epithelial": {
+        "min": 0.4,
+        "max": 0.4,
+        "median": 0.4
+    }
+}

corpus_reference.json

@@ -0,0 +1,289 @@
+{
+ "biomaterials": [
+  "Alginate (%w/v)",
+  "PVA-HA (%w/v)",
+  "CaSO4 (%w/v)",
+  "Na2HPO4 (%w/v)",
+  "Gelatin (%w/v)",
+  "GelMA (%w/v)",
+  "laponite (%w/v)",
+  "graphene oxide (%w/v)",
+  "hydroxyapatite (%w/v)",
+  "Hyaluronic_Acid (%w/v)",
+  "hyaluronan methacrylate (%w/v)",
+  "NorHA (%w/v)",
+  "Fibroin/Fibrinogen (%w/v)",
+  "Pluronic P-123 (%w/v)",
+  "Collagen (%w/v)",
+  "Chitosan (%w/v)",
+  "CS-AEMA (%w/v)",
+  "RGD (mM)",
+  "TCP (%w/v)",
+  "Gellan (%w/v)",
+  "bioactive glass (%w/v)",
+  "Nano/Methylcellulose (%w/v)",
+  "PEGTA (%w/v)",
+  "PEGMA (%w/v)",
+  "PEGDA (%w/v)",
+  "Agarose (%w/v)",
+  "hyaluronic acid+ Ph moieties (%w/v)",
+  "matrigel (%w/v)",
+  "CaCl2 (mM)",
+  "NaCl (mM)",
+  "BaCl2 (mM)",
+  "SrCl2 (mM)",
+  "CaCO3 (mM)",
+  "Genipin (%w/v)",
+  "PVA (%wt)",
+  "trans-glutaminase (%w/v)",
+  "alginate lyase (U/ml)",
+  "D-glucose (%w/v)",
+  "PLGA (%w/v)",
+  "vascular tissue-derived dECM (%w/v)",
+  "PEG-8-SH (mM)",
+  "Alginate dialdehyde (%w/v)",
+  "Alginate sulfate (%w/v)",
+  "RGD-modified alginate (%w/v)",
+  "poly(N-isopropylacrylamide) grafted hyaluronan (%w/v)",
+  "chondroitin sulfate methacrylate (%w/v)",
+  "PCL (%w/v)",
+  "alginate methacrylate (%w/v)",
+  "HRP (U/ml)",
+  "Pluronic F127 / Lutrol F127 (%w/v)",
+  "Irgacure 2959 (%w/v)",
+  "Eosin Y (%w/v)",
+  "Ruthenium (mM)",
+  "sodium persulfate (SPS) (mM)",
+  "HEPES (mM)",
+  "LAP (%w/v)",
+  "glutaraldehyde (%w/v)",
+  "PBS (M)",
+  "glycerol (%w/v)",
+  "cECM (%w/v)",
+  "gel-fu (%w/v)",
+  "Rose Bengal (%w/v)",
+  "Vitamin B2 (%w/v)",
+  "VEGF (%w/v)",
+  "Polypyrrole:PSS (%w/v)",
+  "borate bioactive glass (%w/v)",
+  "astaxanthin (%w/v)",
+  "PRP (%v/v)",
+  "methacrylated collagen (%w/v)",
+  "\u03b1-Toc (\u00b5M)",
+  "ascorbic acid (mM)",
+  "Liver dECM (%w/v)",
+  "galactosylated alginate (%w/v)",
+  "SC-PEG (%w/v)",
+  "SFMA-L (%w/v)",
+  "SFMA-M (%w/v)",
+  "SFMA-H (%w/v)",
+  "KdECMMA (%w/v)",
+  "BA silk fibroin (%w/v)",
+  "Carrageenan (%v)",
+  "Carbopol ETD 2020 NF (%w/v)",
+  "Carbopol Ultrez 10 NF (%w/v)",
+  "Carbopol NF-980 (%w/v)",
+  "FBS (%v/v)",
+  "MeTro (%w/v)",
+  "Triethanolamine (%v/v)",
+  "PEG-Fibrinogen (%w/v)",
+  "polyethylene glycol dimethacrylate (%w/v)",
+  "aprotinin (\u00b5g/ml)",
+  "gold nanorod (mg/mL)",
+  "egg white (w/v)",
+  "1-Vinyl-2-pyrrolidone (v/v)",
+  "carboxyl functionalized carbon nanotubes (%w/v)",
+  "polyHIPE (%w/v)",
+  "\u03b2-D galactose (mM)",
+  "hydrogen peroxide (H2O2) (%v/v)",
+  "lactic acid (v/v)",
+  "NorCol (%w/v)",
+  "DTT (%w/v)",
+  "ammonium persulfate (mM)",
+  "diTyr-RGD (mM)",
+  "PHEG-Tyr (%w/v)",
+  "MMP2-degradable peptide (%w/v)",
+  "KdECM (%w/v)",
+  "EDC (mg)",
+  "NHS (mg)",
+  "VA086 (%w/v)",
+  "PGS (%w/v)",
+  "thiolated HA (%w/v)",
+  "boron nitride nanotubes (%w/v)",
+  "PEDOT:PSS (ul)",
+  "KCl (mM)",
+  "skeletal muscle ECM methacrylate (%w/v)",
+  "PEO (%w/v)",
+  "Carbon dots (mg/ml)",
+  "Laminin (ug/ml)",
+  "DF-PEG (%w/v)",
+  "omentum ECM (%w/v)",
+  "thrombin (unit/ml)",
+  "Carbon nanotube (CNT) (w/v)",
+  "Phytagel (%v)",
+  "Laponite-XLG (%w/w)",
+  "sodium carboxymethyl cellulose (mg)",
+  "lysozyme amyloid nanofibrils:gold nanoparticles (mg)",
+  "Lysozyme amyloid nanofibrils (mg)",
+  "rGO (mg/ml)",
+  "Methacrylated gellan gum (%w/v)",
+  "Acetylsalicylic Acid (%w/w)",
+  "PVA methacrylate (%w/v)",
+  "MXene (mg/ml)"
+ ],
+ "print_params": [
+  "Physical Crosslinking Duration (s)",
+  "Photo Crosslinking Duration (s)",
+  "Extrusion Pressure (kPa)",
+  "Nozzle Movement Speed (mm/s)",
+  "Nozzle Diameter (\u00b5m)",
+  "Syringe Temperature (\u00b0C)",
+  "Substrate Temperature (\u00b0C)"
+ ],
+ "predictors": [
+  "Alginate (%w/v)",
+  "PVA-HA (%w/v)",
+  "CaSO4 (%w/v)",
+  "Na2HPO4 (%w/v)",
+  "Gelatin (%w/v)",
+  "GelMA (%w/v)",
+  "laponite (%w/v)",
+  "graphene oxide (%w/v)",
+  "hydroxyapatite (%w/v)",
+  "Hyaluronic_Acid (%w/v)",
+  "hyaluronan methacrylate (%w/v)",
+  "NorHA (%w/v)",
+  "Fibroin/Fibrinogen (%w/v)",
+  "Pluronic P-123 (%w/v)",
+  "Collagen (%w/v)",
+  "Chitosan (%w/v)",
+  "CS-AEMA (%w/v)",
+  "RGD (mM)",
+  "TCP (%w/v)",
+  "Gellan (%w/v)",
+  "bioactive glass (%w/v)",
+  "Nano/Methylcellulose (%w/v)",
+  "PEGTA (%w/v)",
+  "PEGMA (%w/v)",
+  "PEGDA (%w/v)",
+  "Agarose (%w/v)",
+  "hyaluronic acid+ Ph moieties (%w/v)",
+  "matrigel (%w/v)",
+  "CaCl2 (mM)",
+  "NaCl (mM)",
+  "BaCl2 (mM)",
+  "SrCl2 (mM)",
+  "CaCO3 (mM)",
+  "Genipin (%w/v)",
+  "PVA (%wt)",
+  "trans-glutaminase (%w/v)",
+  "alginate lyase (U/ml)",
+  "D-glucose (%w/v)",
+  "PLGA (%w/v)",
+  "vascular tissue-derived dECM (%w/v)",
+  "PEG-8-SH (mM)",
+  "Alginate dialdehyde (%w/v)",
+  "Alginate sulfate (%w/v)",
+  "RGD-modified alginate (%w/v)",
+  "poly(N-isopropylacrylamide) grafted hyaluronan (%w/v)",
+  "chondroitin sulfate methacrylate (%w/v)",
+  "PCL (%w/v)",
+  "alginate methacrylate (%w/v)",
+  "HRP (U/ml)",
+  "Pluronic F127 / Lutrol F127 (%w/v)",
+  "Irgacure 2959 (%w/v)",
+  "Eosin Y (%w/v)",
+  "Ruthenium (mM)",
+  "sodium persulfate (SPS) (mM)",
+  "HEPES (mM)",
+  "LAP (%w/v)",
+  "glutaraldehyde (%w/v)",
+  "PBS (M)",
+  "glycerol (%w/v)",
+  "cECM (%w/v)",
+  "gel-fu (%w/v)",
+  "Rose Bengal (%w/v)",
+  "Vitamin B2 (%w/v)",
+  "VEGF (%w/v)",
+  "Polypyrrole:PSS (%w/v)",
+  "borate bioactive glass (%w/v)",
+  "astaxanthin (%w/v)",
+  "PRP (%v/v)",
+  "methacrylated collagen (%w/v)",
+  "\u03b1-Toc (\u00b5M)",
+  "ascorbic acid (mM)",
+  "Liver dECM (%w/v)",
+  "galactosylated alginate (%w/v)",
+  "SC-PEG (%w/v)",
+  "SFMA-L (%w/v)",
+  "SFMA-M (%w/v)",
+  "SFMA-H (%w/v)",
+  "KdECMMA (%w/v)",
+  "BA silk fibroin (%w/v)",
+  "Carrageenan (%v)",
+  "Carbopol ETD 2020 NF (%w/v)",
+  "Carbopol Ultrez 10 NF (%w/v)",
+  "Carbopol NF-980 (%w/v)",
+  "FBS (%v/v)",
+  "MeTro (%w/v)",
+  "Triethanolamine (%v/v)",
+  "PEG-Fibrinogen (%w/v)",
+  "polyethylene glycol dimethacrylate (%w/v)",
+  "aprotinin (\u00b5g/ml)",
+  "gold nanorod (mg/mL)",
+  "egg white (w/v)",
+  "1-Vinyl-2-pyrrolidone (v/v)",
+  "carboxyl functionalized carbon nanotubes (%w/v)",
+  "polyHIPE (%w/v)",
+  "\u03b2-D galactose (mM)",
+  "hydrogen peroxide (H2O2) (%v/v)",
+  "lactic acid (v/v)",
+  "NorCol (%w/v)",
+  "DTT (%w/v)",
+  "ammonium persulfate (mM)",
+  "diTyr-RGD (mM)",
+  "PHEG-Tyr (%w/v)",
+  "MMP2-degradable peptide (%w/v)",
+  "KdECM (%w/v)",
+  "EDC (mg)",
+  "NHS (mg)",
+  "VA086 (%w/v)",
+  "PGS (%w/v)",
+  "thiolated HA (%w/v)",
+  "boron nitride nanotubes (%w/v)",
+  "PEDOT:PSS (ul)",
+  "KCl (mM)",
+  "skeletal muscle ECM methacrylate (%w/v)",
+  "PEO (%w/v)",
+  "Carbon dots (mg/ml)",
+  "Laminin (ug/ml)",
+  "DF-PEG (%w/v)",
+  "omentum ECM (%w/v)",
+  "thrombin (unit/ml)",
+  "Carbon nanotube (CNT) (w/v)",
+  "Phytagel (%v)",
+  "Laponite-XLG (%w/w)",
+  "sodium carboxymethyl cellulose (mg)",
+  "lysozyme amyloid nanofibrils:gold nanoparticles (mg)",
+  "Lysozyme amyloid nanofibrils (mg)",
+  "rGO (mg/ml)",
+  "Methacrylated gellan gum (%w/v)",
+  "Acetylsalicylic Acid (%w/w)",
+  "PVA methacrylate (%w/v)",
+  "MXene (mg/ml)",
+  "Cell Line",
+  "Cell Density (million cells/mL)",
+  "Physical Crosslinking Duration (s)",
+  "Photo Crosslinking Duration (s)",
+  "Extrusion Pressure (kPa)",
+  "Nozzle Movement Speed (mm/s)",
+  "Nozzle Diameter (\u00b5m)",
+  "Syringe Temperature (\u00b0C)",
+  "Substrate Temperature (\u00b0C)"
+ ],
+ "targets": [
+  "Printability",
+  "Cell Response"
+ ],
+ "n_records": 2646
+}

corpus_reference.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:93af4ac5ab671ead33c91de0ebcc7888b08c1d8b41abb217445f5404a609d584
+size 116700

deploy/models/classifiers/random/cell_response__adaboost.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8933115563a17d4ba1ce2bcb15daa1564c86e5b697a11a3350c8424fef507f93
+size 44341

deploy/models/classifiers/random/cell_response__bagged_trees.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ec7fd79460aca6c9212d7757da8b19b01f50ef17b35fbb1e4817a935429fd8f3
+size 1306142

deploy/models/classifiers/random/cell_response__bernoulli_naive_bayes.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:282feda9b6e91e96d5dd1ec9cf6ec43d51e8658a38aae0ff94f763ee9e800368
+size 4218

deploy/models/classifiers/random/cell_response__catboost.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e8acf6b8932614de2bd91250155ade47fe8a9249d4190d24aacb7212f8ec9a9f
+size 110986

deploy/models/classifiers/random/cell_response__decision_tree.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fec56bf6dad067b23827be89f8a816e5c10871268acce82da80894fbb58a5e25
+size 3478

deploy/models/classifiers/random/cell_response__extra_tree.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0128ac01ef824a6b8df6f745106d695b6a593a292666ae517d3d68924d514153
+size 3469

deploy/models/classifiers/random/cell_response__extra_trees.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2a1ebaf96b9d13fcc91b5a3c1ca09633eab16f7b33162624917e435cde2f2efd
+size 1600493

deploy/models/classifiers/random/cell_response__gaussian_naive_bayes.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d00e8cc7debd29d176894f559c3b043bd0ef03c6b1d0ecc6c86fa2149f6855d0
+size 7337

deploy/models/classifiers/random/cell_response__gradient_boosting.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0a4332f43c8af0932c05662b97fde1f0d67525773457cd72ddcc670273462767
+size 28636

deploy/models/classifiers/random/cell_response__hist_gradient_boosting.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2bf42d2ad81b2b130b6d295e9d97480d9ada674e89d701575c2bf7be6d379ecb
+size 645934

deploy/models/classifiers/random/cell_response__k_nearest_neighbours.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:780fbe925315a38f89bbf907dac7e4b1f587266533df7c0115e0fc04276b40a0
+size 53470

deploy/models/classifiers/random/cell_response__k_nn_distance_weighted.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:534a360ee86a358052a823d1107c6c002e4e9ef58a3e84f1c2c860f025c2222d
+size 127106

deploy/models/classifiers/random/cell_response__lightgbm.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2c198a22bf0a01258f0e9196472c74797ae0529ab244b1c3e0fe0d6c40e1d23a
+size 552411

deploy/models/classifiers/random/cell_response__linear_discriminant.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:14d6b11c6b801eef3e7f9fa0c6ad28c0a9366fad15cce208bd00725d313adcb2
+size 145025

deploy/models/classifiers/random/cell_response__linear_svm.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6f41c3e869a8250cbbf74ea42271cd3a07684d46794b48a1ff042f29cf8cd9d9
+size 5535

deploy/models/classifiers/random/cell_response__logistic_regression.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fe67a6655375a6e0bfcc706115cc992ed566ba87e57a5696f22843b1b681c1a9
+size 6535

deploy/models/classifiers/random/cell_response__logistic_regression_balanced.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fc90e1018ef9d648de7adb5da1b60e267f4b1e357f538b8fdaa5c6b246c0c97d
+size 6514

deploy/models/classifiers/random/cell_response__mlp_256_128.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2f4bf03e58aea3bb7c171b22d4bf03077652236328dcbf30317765bd28c95a35
+size 877150

deploy/models/classifiers/random/cell_response__nearest_centroid.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:579188d8f4730741cc6d4fcec2d042de26c5f1c4e9810ab23851e2152ec467e3
+size 7171

deploy/models/classifiers/random/cell_response__passive_aggressive.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6d8c577687544c0044204267fce8e94a8fcd695be0e7d8e4660e029bf2b409f7
+size 12296

deploy/models/classifiers/random/cell_response__perceptron.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:12bd23ecb4a5e58d8a3ba47a549bc38b7e9d027a1dde92c474a5e673b8cd3d34
+size 4111

deploy/models/classifiers/random/cell_response__quadratic_discriminant.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:006ecba15ba6131d1dff15163c6168615363be3716e8eb3fc96a4b63b0f44fc7
+size 29521

deploy/models/classifiers/random/cell_response__random_forest.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1b5b4c7ddd69ecf0575e2cad3dc3b94992ded2934f1fbc74498182984e396165
+size 4994

deploy/models/classifiers/random/cell_response__random_forest_balanced.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7be9e347e9d4921977931954bf9d2326544452d96d7bcd2e9ddbfae163a95b77
+size 3902841

deploy/models/classifiers/random/cell_response__ridge_classifier.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:887429d6d2f49ef7c671d9be75c29bb94bc1c7e3abbe202dba13596533c96af9
+size 6651

deploy/models/classifiers/random/cell_response__sgd_hinge.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0070b38894cb7ef91f72dae63d7efe23ce835ee369e481d3b366062c886d3baa
+size 4886

deploy/models/classifiers/random/cell_response__soft_voting_rf_xgb_lr.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:19f3ea953fb061519b7d0fa42416c8edacffc93f762e87cab7a90f5692643340
+size 5233168

deploy/models/classifiers/random/cell_response__stacking_rf_xgb_lr_lr.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6ed798ca8edf3a2f54aae8fde0cc0aaa3eba4457897b97ffa19be49c85c035ba
+size 5233839

deploy/models/classifiers/random/cell_response__svm_polynomial.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bd644c182976274d6ec346169b332e3f18d29912c2152e91f3b6ae17c3155a34
+size 60088

deploy/models/classifiers/random/cell_response__svm_rbf.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ec357aa39dbb6631b933c1e98e4d4b2df6135cc5237a40b2d24dfb6c7d128e45
+size 66004

deploy/models/classifiers/random/cell_response__xgboost.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e9408add167a4c715a875fd88daf4457c46f058ba91c8fa6f9ca57ce1b7d9bb8
+size 189666

deploy/models/classifiers/random/printability__adaboost.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:61fa142eee65a9a3dab6e40de5ec7715340c56563254ca52623e45f0a3681625
+size 48212

deploy/models/classifiers/random/printability__bagged_trees.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:98d12f997b7e7bcb6789ddbf32dacddcc53c859e265778f18682b7b693397e51
+size 3775738

deploy/models/classifiers/random/printability__bernoulli_naive_bayes.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:98d82bc7b87ea7a6e51932768636f4f9cb2e62d7f796d1f1f2aa420230ae0071
+size 3497

deploy/models/classifiers/random/printability__catboost.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:45c920875ad42ed56cade9561ab095e950e209819d68f2e061ff68534dd73fd3
+size 787008

deploy/models/classifiers/random/printability__decision_tree.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:768ec1872e1fc323154af785e6f71d554389de56fb43aa2c0624e6a70852a6b5
+size 10029

deploy/models/classifiers/random/printability__extra_tree.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:591f152e86031f04d9d92a5cd88164dcb636509e63d6726010c6814b512bfa75
+size 9601

deploy/models/classifiers/random/printability__extra_trees.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e152e1e7423d5e3306b7abe26ee13ed67596433b1e910f818ed6840a284fffa7
+size 487961

deploy/models/classifiers/random/printability__gaussian_naive_bayes.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d806b6407b0cc8ab12d7fa836dee8811e349c8e740b7983016111b02d2239223
+size 6142

deploy/models/classifiers/random/printability__gradient_boosting.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:47a1931819018e107c00db7279b2e46722b92606c7d7fc1bf95f5f2728e9e7fa
+size 955239

deploy/models/classifiers/random/printability__hist_gradient_boosting.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8503f4f620cc539db9e017106b8df556958c3e3191c60da4568d428ac0e2f64f
+size 1555799

deploy/models/classifiers/random/printability__k_nearest_neighbours.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:454d01feb6c9b0927614bf2537641b445aac81664bb0df17b33a7472fccceb49
+size 53695

deploy/models/classifiers/random/printability__k_nn_distance_weighted.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:09f59127db7f4e8d30b7ad810bcdd3eda19749fb13b10ac68fe4e14a32991767
+size 53689