gbyuvd/chemq3-molsim-sft-smiles

ECFP4 Molecular Pairs Dataset

A dataset of molecular pairs with ECFP4 Dice similarity scores uniformly sampled across a target range, using FAISS for efficient similarity search. This pipeline intended to generate a high-quality dataset of molecular pairs for similarity-based learning, balancing chemical diversity, computational efficiency, and target similarity distribution. Specially designed to retain only pairs with 0.5 ≤ Dice(MACCS) ≤ 0.95—a targeted range for supervised fine-tuning (SFT) and sentence-transformers training aimed at learning meaningful but non-trivial molecular similarities.

🎯 Objective

Produce a balanced set of molecular pairs where the Dice similarity (based on ECFP4 fingerprints) falls within a specified range (e.g., 0.5–0.95), with approximately equal representation across similarity bins.

📦 Input

• comb_smi.csv: CSV file containing a column SMILES with input molecules.
• the dataset is curated and combined from ChemBL34, COCONUTDB, and SuperNatural3

⚙️ Key Steps

1. Preprocessing:
   • Remove salts (keep largest fragment).
   • Canonicalize SMILES and deduplicate.
2. Fingerprinting:
   • Compute ECFP4 (Morgan radius=2, 2048-bit folded) fingerprints using RDKit.
3. Indexing:
   • Build a FAISS IndexFlatIP for fast inner-product (bitwise intersection) search.
4. Pair Sampling:
   • For each molecule, retrieve nearest neighbors.
   • Compute Dice similarity: ( \text{Dice} = \frac{2 \cdot |A \cap B|}{|A| + |B|} ).
   • Assign pairs to bins within [0.5, 0.95] and sample up to 200,000 pairs per bin.
5. Output:
   • Save pairs as pairs_ecfp4.parquet (columns: mol1, mol2, sim).
   • Generate and save a histogram of similarity scores (_histogram.png and .pdf).

📁 Output Files

• pairs_ecfp4.parquet: Final dataset of molecular pairs with similarity scores.
• pairs_ecfp4_histogram.png / .pdf: Visualization of similarity distribution and binning.

⚠️ Notes

• Designed for large-scale datasets; uses batching and memory-efficient FAISS search.
• Default configuration processes all molecules; set N_MOLS for testing.
• Only valid, unique, canonical SMILES are retained.
• Due to compute constraints I am unable to generate more samples

📦 Requirements

• Python 3.8+
• pandas, numpy, faiss-cpu, rdkit, tqdm, matplotlib, seaborn

Citations

ChEMBL34:

@misc{chembl34,
  title={ChemBL34},
  year={2023},
  doi={10.6019/CHEMBL.database.34}
}

@article{zdrazil2023chembl,
  title={The ChEMBL Database in 2023: a drug discovery platform spanning multiple bioactivity data types and time periods},
  author={Zdrazil, Barbara and Felix, Eloy and Hunter, Fiona and Manners, Emma J and Blackshaw, James and Corbett, Sybilla and de Veij, Marleen and Ioannidis, Harris and Lopez, David Mendez and Mosquera, Juan F and Magarinos, Maria Paula and Bosc, Nicolas and Arcila, Ricardo and Kizil{\"o}ren, Tevfik and Gaulton, Anna and Bento, A Patr{\'i}cia and Adasme, Melissa F and Monecke, Peter and Landrum, Gregory A and Leach, Andrew R},
  journal={Nucleic Acids Research},
  year={2023},
  volume={gkad1004},
  doi={10.1093/nar/gkad1004}
}

COCONUTDB:

@article{sorokina2021coconut,
  title={COCONUT online: Collection of Open Natural Products database},
  author={Sorokina, Maria and Merseburger, Peter and Rajan, Kohulan and Yirik, Mehmet Aziz and Steinbeck, Christoph},
  journal={Journal of Cheminformatics},
  volume={13},
  number={1},
  pages={2},
  year={2021},
  doi={10.1186/s13321-020-00478-9}
}

SuperNatural3:

@article{Gallo2023,
  author = {Gallo, K and Kemmler, E and Goede, A and Becker, F and Dunkel, M and Preissner, R and Banerjee, P},
  title = {{SuperNatural 3.0-a database of natural products and natural product-based derivatives}},
  journal = {Nucleic Acids Research},
  year = {2023},
  month = jan,
  day = {6},
  volume = {51},
  number = {D1},
  pages = {D654-D659},
  doi = {10.1093/nar/gkac1008}
}