Datasets for P2Rank project

These are datasets used by P2Rank ligand binding site prediction tool for training and evaluation.

Each *.ds file contains a list of items that form a dataset with actual data being stored in subdirectories.

! Important note !

Please note that *.ds files that define the datasets from our papers may contain only subsets of PDB files in individual directories. For example holo4k/ directory contains 4543 pdb files but holo4k.ds contains 4009 lines. For reproducibility, 4009 is the correct number of proteins in the HOLO4K dataset used in P2Rank/PrankWeb papers.

Datasets

Main sets of proteins:

• CHEN11: a dataset of 251 proteins harboring 476 ligands introduced in LBS prediction benchmarking study
• ASTEX: Astex Diverse set
• metapocket2 datasets
  • U/B48: Datasets that contain a set of 48 proteins in a bound and unbound state
  • DT198: a dataset of 198 drug-target complexes
  • B210: a benchmarking dataset of 210 proteins in a bound state
• FPTRAIN: dataset used by Fpocket for training its pocket scoring function
• HOLO4K: a large dataset of protein-ligand complexes. Contains larger multi-chain structures downloaded directly from PDB. Disjunct with CHEN11 and JOINED.

Variations

• "standard" ... 1 column of liganated proteins
• *(mlig)* datasets: datasets that contain explicitly specified relevant ligands. Valid ligand codes come from MOAD 2013 database. Proteins unknown to MOAD and proteins with conflicting ligand codes (valid&invalid) were removed.
• datasets with predictions: include predictions by other ligand binding site prediction methods (-fpocket.ds, -sitehound.ds, etc. suffixes)
• *-XXsubset-* datasets: contain a subset of the original datasets for which the given method finished successfully and produced predictions (mp: MetaPocket2, sh: SiteHound, ds: DeepSite)

Predictions

This repository also contains binding site predictions produced by some other methods.

• Fpocket
  • used version: v1.0 with default parameters
• SiteHound
  • used version: version labeled as
  • command used to generate predictions: ls *.pdb | xargs -i python ../auto.py -i {} -p CMET -k (executed in a directory with pdb files)
  • default probe and parameters were used
• MetaPocket 2.0
  • obtained from MetaPocket 2.0 web server by a Python script in Fall 2017 using default parameters
• DeepSite
  • obtained from DeepSite web server by a Python script in Fall 2017 using default parameters
• P2Rank
  • correspond to P2Rank 2.0 with default parameters

modifications

• 1xgf.pdb removed from holo4k datasets (all UNK groups, no ligands)