README.md

---
license: cc-by-sa-4.0
task_categories:
  - text-classification
  - text-generation
  - tabular-classification
language:
  - en
size_categories:
  - 10M<n<100M
tags:
  - biology
  - chemistry
  - drug-discovery
  - clinical-trials
  - protein-protein-interaction
  - gene-essentiality
  - negative-results
  - publication-bias
  - benchmark
  - biomedical
pretty_name: "NegBioDB: Negative Results Database & Benchmark"
configs:
  - config_name: dti_pairs
    data_files: "data/negbiodb_dti_pairs.parquet"
  - config_name: dti_m1_balanced
    data_files: "data/negbiodb_m1_balanced.parquet"
  - config_name: ct_pairs
    data_files: "data/ct/negbiodb_ct_pairs.parquet"
  - config_name: ppi_pairs
    data_files: "data/ppi/negbiodb_ppi_pairs.parquet"
  - config_name: ge_pairs
    data_files: "data/ge/negbiodb_ge_pairs.parquet"
---

# NegBioDB

**Negative Results Database & Dual ML/LLM Benchmark for Biomedical Sciences**

[![License: CC BY-SA 4.0](https://img.shields.io/badge/License-CC_BY--SA_4.0-lightgrey.svg)](https://creativecommons.org/licenses/by-sa/4.0/)
[![Python 3.11+](https://img.shields.io/badge/python-3.11+-blue.svg)](https://www.python.org/downloads/)

Approximately 90% of scientific experiments produce null or inconclusive results, yet the vast majority remain unpublished. NegBioDB systematically collects experimentally confirmed negative results across four biomedical domains and provides dual-track ML + LLM benchmarks to quantify the impact of this publication bias on AI models.

## Quick Start

```python
# Load any domain with the datasets library
from datasets import load_dataset

# DTI pairs (30.5M negative compound-target interactions)
dti = load_dataset("jang1563/NegBioDB", "dti_pairs", split="train")

# Clinical trial failures (102K intervention-condition pairs)
ct = load_dataset("jang1563/NegBioDB", "ct_pairs", split="train")

# PPI negatives (2.2M confirmed non-interactions)
ppi = load_dataset("jang1563/NegBioDB", "ppi_pairs", split="train")

# Gene essentiality (22.5M gene-cell-line pairs)
ge = load_dataset("jang1563/NegBioDB", "ge_pairs", split="train")
```

Or load directly with pandas:

```python
import pandas as pd
from huggingface_hub import hf_hub_download

# Download a specific file
path = hf_hub_download("jang1563/NegBioDB", "data/negbiodb_dti_pairs.parquet", repo_type="dataset")
df = pd.read_parquet(path)
print(df.shape)  # (1_725_446, ~20 columns)
```

## Database Statistics

| Domain | Negative Results | Key Entities | Sources | DB Size |
|--------|-----------------|--------------|---------|---------|
| **DTI** | 30,459,583 | 919K compounds, 3.7K targets | ChEMBL, PubChem, BindingDB, DAVIS | ~21 GB |
| **CT** | 132,925 | 177K interventions, 56K conditions | AACT, CTO, Open Targets, Shi & Du | ~500 MB |
| **PPI** | 2,229,670 | 18.4K proteins | IntAct, HuRI, hu.MAP, STRING | 849 MB |
| **GE** | 28,759,256 | 19,554 genes, 2,132 cell lines | DepMap (CRISPR, RNAi) | ~16 GB |
| **Total** | **~61.6M** | | **14 sources** | **~38 GB** |

## Key Findings

### ML: Negative Source Matters

**DTI** -- Degree-matched negatives inflate LogAUC by +0.112 on average. Cold-target splits cause catastrophic failure:

| DTI Model | Random (NegBioDB) | Random (Degree-Matched) | Cold-Target |
|-----------|------------------|------------------------|-------------|
| DeepDTA | 0.833 | **0.919** | 0.325 |
| GraphDTA | 0.843 | **0.967** | 0.241 |
| DrugBAN | 0.830 | **0.955** | 0.151 |

**PPI** -- PIPR cold_both AUROC drops to 0.409 (below random); MLPFeatures remains robust at 0.950.

**CT** -- NegBioDB negatives are trivially separable (AUROC ~1.0); M2 7-way classification is challenging (best macro-F1 = 0.51).

**GE** -- Cold-gene splits reveal severe generalization gaps; degree-balanced negatives modestly improve ranking metrics.

### LLM: L4 Discrimination Reveals Domain Differences

| Domain | L4 MCC Range | Interpretation | Contamination |
|--------|-------------|----------------|---------------|
| DTI | &le; 0.18 | Near random | Not detected |
| PPI | 0.33--0.44 | Moderate | **Yes** (temporal gap) |
| CT | 0.48--0.56 | Meaningful | Not detected |
| GE | Pending | -- | -- |

## Exported Datasets

### DTI

| File | Description |
|------|-------------|
| `negbiodb_dti_pairs.parquet` | 1.7M compound-target pairs with 5 split columns |
| `negbiodb_m1_balanced.parquet` | M1: 1.73M rows (1:1 active:inactive) |
| `negbiodb_m1_realistic.parquet` | M1: 9.49M rows (1:10 ratio) |
| `negbiodb_m1_balanced_ddb.parquet` | Degree-balanced split |
| `negbiodb_m1_uniform_random.parquet` | Uniform random negatives (control) |
| `negbiodb_m1_degree_matched.parquet` | Degree-matched negatives (control) |

### CT

| File | Description |
|------|-------------|
| `ct/negbiodb_ct_pairs.parquet` | 102,850 failure pairs, 6 splits |
| `ct/negbiodb_ct_m1_balanced.parquet` | Binary: 11,222 rows (5,611 pos + 5,611 neg) |
| `ct/negbiodb_ct_m2.parquet` | 7-way category: 112,298 rows |

### PPI

| File | Description |
|------|-------------|
| `ppi/negbiodb_ppi_pairs.parquet` | 2,220,786 negative pairs with split columns |
| `ppi/ppi_m1_balanced.parquet` | M1: 123,456 rows (1:1 pos:neg) |
| `ppi/ppi_m1_realistic.parquet` | M1: 679,008 rows (1:10 ratio) |

### GE

| File | Description |
|------|-------------|
| `ge/negbiodb_ge_pairs.parquet` | 22.5M gene-cell-line pairs with 5 split columns |

### LLM Benchmarks

| File | Description |
|------|-------------|
| `llm_benchmarks/l1_mcq.jsonl` | L1: Multiple-choice classification |
| `llm_benchmarks/l4_tested_untested.jsonl` | L4: Tested vs. untested discrimination |
| `ct_llm/ct_l*_dataset.jsonl` | CT domain LLM datasets (L1-L4) |
| `ppi_llm/ppi_l*_dataset.jsonl` | PPI domain LLM datasets (L1-L4) |
| `ge_llm/ge_l*_dataset.jsonl` | GE domain LLM datasets (L1-L4) |

## Data Sources

| Source | Records | License |
|--------|---------|---------|
| [ChEMBL v36](https://www.ebi.ac.uk/chembl/) | 371K | CC BY-SA 3.0 |
| [PubChem BioAssay](https://pubchem.ncbi.nlm.nih.gov/) | 29.6M | Public Domain |
| [BindingDB](https://www.bindingdb.org/) | 404K | CC BY |
| [DAVIS](https://github.com/dingyan20/Davis-Dataset-for-DTA-Prediction) | 20K | Public |
| [AACT (ClinicalTrials.gov)](https://aact.ctti-clinicaltrials.org/) | 216,987 trials | Public Domain |
| [CTO](https://github.com/fairnessforensics/CTO) | 20,627 | MIT |
| [Open Targets](https://www.opentargets.org/) | 32,782 targets | Apache 2.0 |
| [Shi & Du 2024](https://doi.org/10.1038/s41597-024-03399-2) | 119K + 803K rows | CC BY 4.0 |
| [IntAct](https://www.ebi.ac.uk/intact/) | 779 pairs | CC BY 4.0 |
| [HuRI](http://www.interactome-atlas.org/) | 500,000 pairs | CC BY 4.0 |
| [hu.MAP 3.0](https://humap3.proteincomplexes.org/) | 1,228,891 pairs | MIT |
| [STRING v12.0](https://string-db.org/) | 500,000 pairs | CC BY 4.0 |
| [DepMap CRISPR](https://depmap.org/) | 28.7M gene-cell pairs | CC BY 4.0 |
| [DepMap RNAi](https://depmap.org/) | Integrated | CC BY 4.0 |

## Reproducing from Source

Full pipeline code is available at [GitHub](https://github.com/jang1563/NegBioDB).

```bash
git clone https://github.com/jang1563/NegBioDB.git
cd NegBioDB
make setup          # Create venv and install dependencies
make db             # Initialize SQLite database
make download       # Download all sources
make load-all       # Run ETL pipelines
```

## Citation

```bibtex
@misc{negbiodb2026,
  title={NegBioDB: A Negative Results Database and Dual ML/LLM Benchmark for Biomedical Sciences},
  author={Kim, JangKeun},
  year={2026},
  url={https://github.com/jang1563/NegBioDB}
}
```

## License & Contact

**License:** CC BY-SA 4.0 (required by ChEMBL's CC BY-SA 3.0 viral clause)

**Contact:** JangKeun Kim (jak4013@med.cornell.edu) -- Weill Cornell Medicine