Update Readme.md

README.md

@@ -1,3 +1,131 @@
----
-license: mit
----
+---
+license: mit
+---
+
+# ReverseLigQ Datasets
+
+This repository contains the curated datasets used by **ReverseLigQ**, a platform designed to search for potential protein targets (molecular targets) of bioactive compounds.
+
+The data stored here includes ligand representations, mappings between internal indices and compound identifiers, and curated associations between ligands and Pfam domains derived from structural and sequence information.
+
+---
+
+## Contents
+
+### 1. Embeddings and Fingerprints
+
+- **`comps_embs.npy`**  
+  NumPy array containing **ChemBERTa embeddings** for ligands derived from **PDB** and **ChEMBL**.  
+  Each row corresponds to a ligand, and the position in the array is linked to the internal index (`idx`) used throughout ReverseLigQ.
+
+- **`comps_fps.npy`**  
+  NumPy array containing **Morgan fingerprints** (circular fingerprints, ECFP-like) for the same set of ligands from **PDB** and **ChEMBL**.  
+  As with `comps_embs.npy`, rows are aligned with the internal ligand index (`idx`).
+
+---
+
+### 2. Index–Identifier Mappings
+
+- **`id_to_idx.pkl`**  
+  Python dictionary mapping **ligand identifiers** (e.g., from PDB or ChEMBL) to the **internal index** (`idx`) used in the arrays `comps_embs.npy` and `comps_fps.npy`.  
+  This mapping allows you to go from an external ligand ID to the corresponding row in the embedding and fingerprint matrices.
+
+- **`idx_to_id.pkl`**  
+  Inverse mapping of `id_to_idx.pkl`.  
+  It is a Python dictionary mapping the internal index (`idx`) back to the **original ligand identifier**.  
+  This is useful, for example, when retrieving nearest neighbors in embedding space by index and then needing to recover the original IDs.
+
+---
+
+### 3. Ligand–Pfam Domain Associations
+
+- **`ligs_fams_curated.pkl`**  
+  Python dictionary mapping ligands to **Pfam domains they are known to bind**, based on **curated, high-confidence evidence**.  
+  This curated set includes ligands whose associated domain can be determined unambiguously from:
+  - a **resolved 3D structure** (e.g., a PDB complex where the domain is clearly defined),
+  - a **protein with a single domain**, or
+  - **strong structural similarity** to another ligand whose interaction has been experimentally confirmed in a solved 3D structure.  
+  This file represents the **high-confidence (curated) ligand–domain associations** used by ReverseLigQ.
+
+- **`ligs_fams_possible.pkl`**  
+  Python dictionary mapping ligands to **possible Pfam domains they may bind**, in cases where the specific domain cannot be determined with certainty.  
+  Here, the reference protein:
+  - often contains **multiple domains**,
+  - does **not** have a resolved 3D structure for the ligand–protein complex, and therefore  
+  - it is unclear which of the possible domains is the true binding site.  
+  This file therefore represents **putative or ambiguous ligand–domain associations**, complementing the curated set.
+
+---
+
+## Usage Notes
+
+- The **internal index** (`idx`) is the key that ties together:
+  - rows in `comps_embs.npy`
+  - rows in `comps_fps.npy`
+  - entries in `id_to_idx.pkl` / `idx_to_id.pkl`
+- The **Pfam-domain dictionaries** (`ligs_fams_curated.pkl` and `ligs_fams_possible.pkl`) can be used to:
+  - build training or evaluation sets for ligand–target or ligand–domain prediction,
+  - separate high-confidence from ambiguous associations,
+  - benchmark methods that infer protein targets from ligand similarity.
+
+---
+
+## Example (Python)
+
+```python
+import numpy as np
+import pickle
+
+# Load embeddings and fingerprints
+embs = np.load("comps_embs.npy")
+fps = np.load("comps_fps.npy")
+
+# Load mappings
+with open("id_to_idx.pkl", "rb") as f:
+    id_to_idx = pickle.load(f)
+
+with open("idx_to_id.pkl", "rb") as f:
+    idx_to_id = pickle.load(f)
+
+# Load curated ligand–Pfam associations
+with open("ligs_fams_curated.pkl", "rb") as f:
+    ligs_fams_curated = pickle.load(f)
+
+# Example: get embedding for a given ligand ID
+lig_id = "CHEMBL123456"
+idx = id_to_idx[lig_id]
+lig_emb = embs[idx]
+lig_fp = fps[idx]
+```
+
+## Example 2 (Python)
+
+```python
+import pickle
+
+# Load curated and possible ligand–Pfam associations
+with open("ligs_fams_curated.pkl", "rb") as f:
+    ligs_fams_curated = pickle.load(f)
+
+with open("ligs_fams_possible.pkl", "rb") as f:
+    ligs_fams_possible = pickle.load(f)
+
+lig_id = "CHEMBL123456"
+
+# High-confidence binding domains (if available)
+curated_domains = ligs_fams_curated.get(lig_id, [])
+
+# Possible (ambiguous) domains
+possible_domains = ligs_fams_possible.get(lig_id, [])
+
+print("Curated Pfam domains:", curated_domains)
+print("Possible Pfam domains:", possible_domains)
+```
+
+## Citation
+
+If you use these datasets, please cite:
+
+Schottlender G, Prieto JM, Palumbo MC, Castello FA, Serral F, Sosa EJ, Turjanski AG, Martí MA and Fernández Do Porto D (2022).
+From drugs to targets: Reverse engineering the virtual screening process on a proteomic scale.
+Front. Drug. Discov. 2:969983. doi: 10.3389/fddsv.2022.969983