CHEN11, ASTEX, metapocket2 datasets, FPTRAIN, HOLO4K

Dataset Overview

Main Protein Datasets

1. CHEN11: Contains 251 proteins with a total of 476 ligands for LBS prediction benchmarking.
2. ASTEX: Astex diverse collection.
3. metapocket2 dataset series:
   • U/B48: 48 proteins in both bound and unbound states.
   • DT198: 198 drug‑target complexes.
   • B210: Benchmark dataset of 210 bound‑state proteins.
4. FPTRAIN: Dataset used for training the Fpocket pocket‑scoring function.
5. HOLO4K: Large protein‑ligand complex dataset containing multi‑chain structures, non‑overlapping with CHEN11 and JOINED.

Dataset Variants

• "standard": Contains a single column of ligand‑bound proteins.
• *(mlig)* dataset: Explicitly specifies associated ligands; ligand codes are sourced from the MOAD 2013 database.
• Prediction‑included datasets: Contain predictions from other ligand‑binding site prediction methods.
• *-XXsubset-* datasets: Subsets of the original datasets where a specific method succeeded and produced predictions.

Dataset Caveats

• *.ds files may contain only a subset of the PDB files. For example, the holo4k/ directory holds 4,543 PDB files, but holo4k.ds lists 4,009 lines, which is the correct protein count used in the P2Rank/PrankWeb paper for the HOLO4K dataset.
• 1xgf.pdb has been removed from the holo4k dataset (contains only UNK groups and no ligand).