THU-ATOM/casp_data

CASP Dataset

This dataset contains protein structure prediction data compiled from the Critical Assessment of Structure Prediction (CASP) experiments for years 14, 15, and 16. The data is primarily organized for single-chain protein structure prediction testing.

Dataset Overview

The dataset includes experimental data from three CASP competitions:

• CASP14 (2020)
• CASP15 (2022)
• CASP16 (2024)

Directory Structure

casp_data/
├── casp14/
│   ├── sequences.fasta
│   ├── experiment_structures/
│   │   ├── T1024.pdb
│   │   ├── T1030.pdb
│   │   ├── T1032.pdb
│   │   ├── T1034.pdb
│   │   ├── T1038.pdb
│   │   ├── T1046s1.pdb
│   │   ├── T1046s2.pdb
│   │   └── ...
│   └── precomputed_msa/
│       ├── T1024/
│       │   └── non_pairing.a3m
│       ├── T1030/
│       │   └── non_pairing.a3m
│       └── ...
├── casp15/
│   ├── sequences.fasta
│   ├── experiment_structures/
│   │   └── [PDB files]
│   └── precomputed_msa/
│       └── [MSA directories]
└── casp16/
    ├── sequences.fasta
    ├── experiment_structures/
    │   └── [PDB files]
    └── precomputed_msa/
        └── [MSA directories]

File Descriptions

sequences.fasta

Contains all target protein sequences for the respective CASP competition in FASTA format. Each sequence corresponds to a prediction target with identifiers like T1024, H1151s1, etc.

experiment_structures/

Directory containing experimentally determined protein structures in PDB format. These serve as ground truth structures for evaluating prediction accuracy. Files are named according to their CASP target identifiers (e.g., T1024.pdb).

precomputed_msa/

Contains precomputed Multiple Sequence Alignments (MSA) for each target, organized in subdirectories by target ID. Each subdirectory contains:

• non_pairing.a3m: MSA file in A3M format

MSA Generation Details

The Multiple Sequence Alignments were generated using MMseqs2 with ColabFold scripts, searching against the following databases:

• uniref30_2302_db: UniRef30 database
• colabfold_envdb_202108_db: ColabFold environmental database

The MSA generation command used:

--db1 uniref30_2302_db \
--db3 colabfold_envdb_202108_db

Usage

This dataset is suitable for:

• Evaluating deep learning models for protein structure prediction

Data Attribution

All data in this repository is sourced from publicly available databases and competitions:

• CASP (Critical Assessment of protein Structure Prediction): Protein structure prediction targets and evaluation data from the CASP website (https://predictioncenter.org/)
• PDB (Protein Data Bank): Experimental protein structures from the RCSB Protein Data Bank (https://www.rcsb.org/)

We have organized and compiled this data for research purposes. All rights remain with the original data providers and owners. This repository does not claim ownership of any original data - we simply provide an organized collection for convenience.

Original Data Sources:

• CASP competition data: © CASP organization
• PDB structures: © RCSB Protein Data Bank and original depositors

Please cite the original sources when using this data in your research.