Giters

rdk / p2rank-ions-data

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Introduction

This repository contains development datasets for P2Rank ion binding site prediction.

P2Rank repo: https://github.com/rdk/p2rank

Contents

Directories:

  • _incoming original received files before tast/train splits
  • _processing ... contains notes related to prosessing of datasets
  • ZN, MG, ... datasets for individual ions

ZN datasets

Dataset files are based on 2 main datasets: ZN_ppu and ZN_benchmark:

Description

  • ppu: It contains X-ray structures from PDB, with a resolution of 3.0 Å or better, that include at least one (ion) ligand of interest. The dataset has been treated for protein redundancy by applying a sequence identity cutt-off of 30%, and for chain redundancy with UniProt, so as to discard identical chains within a single protein. It has been filtered to exclude structures from the benchmark dataset.
  • benchmark: This dataset has been used in the literature for evaluating previous predictors. It contains a sub-set of ion-binding proteins from the BioLip database, with a pairwise sequence identity below 30% and a length over 50 residues. The dataset collectively includes proteins that bind nine metal ligands.
  • biolip: TODO

Each of them contain _exposed and _buried version.

Description

  • buried: It contains (ion) ligands that are located below the computed solvent accessible surface area of the protein structure that they bind to, and are thus considered inaccessible to the solution.
  • exposed: It contains (ion) ligands that have a non-zero computed solvent accessible surface area, and are thus at least partially exposed to the solution.

There is also _all version for ZN_ppu (and for ZN_benchmark). Note: ZN_ppu_all.ds is not just union of lines from ZN_ppu_buried.ds and ZN_ppu_exposed.ds because a single protein can contain both buried and exposed ions. Consequently, dataset file lines for such protein differ in buried and exposed .ds files.

Main datasets (exposed/buried split)

  • ZN_benchmark_exposed.ds
  • ZN_benchmark_buried.ds
  • ZN_benchmark_all.ds
  • ZN_ppu_exposed.ds
  • ZN_ppu_buried.ds
  • ZN_ppu_all.ds

train/test split of ZN_ppu (split A)

Additionaly there are datasets that represent train/test split of ZN_ppu. First, single random split of ZN_ppu_all.ds was performed: split "A" (train:66%, test:34%):. Then ZN_ppu_exposed.ds and ZN_ppu_buried.ds were split according to subsets of proteins from split A.

  • ZN_ppu_all_Atrain.ds
  • ZN_ppu_all_Atest.ds
  • ZN_ppu_exposed_Atrain.ds: subset of ZN_ppu_buried.ds containing only proteins from ZN_ppu_all_Atrain.ds
  • ZN_ppu_exposed_Atest.ds: subset of ZN_ppu_exposed.ds containing only proteins from ZN_ppu_all_Atest.ds
  • ZN_ppu_buried_Atrain.ds: subset of ZN_ppu_buried.ds containing only proteins from ZN_ppu_all_Atrain.ds
  • ZN_ppu_buried_Atest.ds: subset of ZN_ppu_buried.ds containing only proteins from ZN_ppu_all_Atest.ds

Consequently, it is possible to train and test on different versions (all/exposed/buried) in a single run: e.g. train on ZN_ppu_all_Atrain.ds and test on ZN_ppu_exposed_Atest.ds or ZN_ppu_buried_Atest.ds.

MG datasets

Similarly as in ZN datasets:

  • There are 2 main datasets: MG_ppu and MG_benchmark.
  • Each of them contain _exposed and _buried version.

There is also _all version for MG_ppu (and for MG_benchmark). Note: MG_ppu_all.ds is not just union of lines from MG_ppu_buried.ds and MG_ppu_exposed.ds because a single protein can contain both buried and exposed ions. Consequently, dataset file lines for such protein differ in buried and exposed .ds files.

Main datasets (exposed/buried split)

  • MG_benchmark_exposed.ds
  • MG_benchmark_buried.ds
  • MG_benchmark_all.ds
  • MG_ppu_exposed.ds
  • MG_ppu_buried.ds
  • MG_ppu_all.ds

train/test split of MG_ppu (split A)

Similarly as in ZN datasets: Additionaly there are datasets that represent train/test split of MG_ppu. First, single random split of MG_ppu_all.ds was performed: split "A" (train:66%, test:34%):. Then MG_ppu_exposed.ds and MG_ppu_buried.ds were split according to subsets of proteins from split A.

  • MG_ppu_all_Atrain.ds
  • MG_ppu_all_Atest.ds
  • MG_ppu_exposed_Atrain.ds: subset of MG_ppu_buried.ds containing only proteins from MG_ppu_all_Atrain.ds
  • MG_ppu_exposed_Atest.ds: subset of MG_ppu_exposed.ds containing only proteins from MG_ppu_all_Atest.ds
  • MG_ppu_buried_Atrain.ds: subset of MG_ppu_buried.ds containing only proteins from MG_ppu_all_Atrain.ds
  • MG_ppu_buried_Atest.ds: subset of MG_ppu_buried.ds containing only proteins from MG_ppu_all_Atest.ds

Consequently, it is possible to train and test on different versions (all/exposed/buried) in a single run: e.g. train on MG_ppu_all_Atrain.ds and test on MG_ppu_exposed_Atest.ds or MG_ppu_buried_Atest.ds.

train/test split of MG_ppu (split B)

Split B is renamed split A from Christos.

  • MG_ppu_all_Btrain.ds
  • MG_ppu_all_Btest.ds
  • MG_ppu_exposed_Btrain.ds
  • MG_ppu_exposed_Btest.ds
  • MG_ppu_buried_Btrain.ds
  • MG_ppu_buried_Btest.ds

About