Chiral Ligand Benchmark Dataset for Protein-Ligand Cofolding Evaluation

PLINDER-derived Chiral Ligand Benchmark Dataset for Protein-Ligand Cofolding Evaluation Overview This benchmark dataset enables systematic evaluation of stereochemical accuracy in AI-based protein-ligand complex structure prediction models. The dataset focuses on chiral small molecule ligands and their stereochemical reproduction challenges. Associated Publication "Improving Stereochemical Limitations in Protein-Ligand Complex Structure Prediction"Authors: R. Ishitani, Y. MoriwakiPublication: https://pubs.acs.org/doi/10.1021/acsomega.5c07675Preprint: https://www.biorxiv.org/content/10.1101/2025.03.25.645362v2 Dataset Files 📊 Dataset Metadata (CSV Files) File Description Size plinder_l95.csv Plinder-L95 dataset (~6,600 entries, Tanimoto similarity ≥0.95) plinder_l70.csv Plinder-L70 dataset (~4,700 entries, Tanimoto similarity ≥0.7) CSV Columns: sys_id: PLINDER system identifier pdb_id: Protein Data Bank ID ccd_id: Chemical Component Dictionary code smiles: Ligand SMILES representation release_date: PDB entry release date split: AlphaFold3 training split (before/after 2021-09-30) afterP95L70: Boolean flag for After-P95L70 subset (excludes proteins >95% similar to training) afterP70L70: Boolean flag for After-P70L70 subset (excludes proteins >70% similar to training) 🧬 Multiple Sequence Alignments (MSA Files) File Description plinder_l70_msa.tar.gz Pre-computed MSAs for Plinder-L70 dataset plinder_l95_msa.tar.gz Pre-computed MSAs for Plinder-L95 dataset Format: A3M files named {sys_id}.a3m Generated with: MMSeqs2 via LocalColabFold Purpose: Ensures consistent MSA inputs across different prediction methods Key Features ✅ Chiral-focused: All ligands contain at least one chiral center✅ Temporal splits: Before/After training cutoffs for fair evaluation✅ Similarity filtering: Multiple stringency levels to assess generalization✅ Ready-to-use MSAs: Pre-computed for consistent benchmarking Reference For detailed methodology and selection criteria, see the Methods section of the associated publication. Citation: Ishitani, R. & Moriwaki, Y. Improving Stereochemical Limitations in Protein-Ligand Complex Structure Prediction. ACS Omega (2025). https://doi.org/10.1021/acsomega.5c07675