PDB

Protein Data Bank (PDB) Dataset - Preprocessing

Implementation Setup

• Python==3.7
• Biopython==1.79
• Scipy==1.7.3

Download

The PDB dataset can be download by listing the proteins in the file proteins_all.txt and running the python script download_pdb.py).

python download_pdb.py [--save_dir SAVE_DIR] [--prot_list PROT_LIST]

Where the arguments are:

    --save_dir    Where to save downloaded PDB files. (default=./PDB_files)
    --prot_list  Text file with list of proteins to be downloaded (default=./proteins_all.txt)

Retreive Residue Sequences and Coordinates

python pdb_to_seq_coords.py [-h] [--pdb_dir PDB_DIR] [--save_dir SAVE_DIR]  

Where the arguments are given as:

    --pdb_dir   Protein PDB files directory.
    --save_dir  Where to save retrieved chain sequences and coordinates.

Transform Protein Chains into Graphs

python create_chain_graphs.py [-h] [--cmap_thresh CMAP_THRESH] [--save_dir SAVE_DIR] [--input_dir INPUT_DIR]  

The arguments are:

    --cmap_thresh  Threshold for contact map.
    --save_dir     Where to save generated protein chain graphs.
    --input_dir    Directory containing protein chain sequences and coordinates

Generate Label Related-data

python create_labels_npz.py [--annot_file ANNOT_FILE] [--save_dir SAVE_DIR] 

The arguments are:

 --annot_file ANNOT_FILE 
 --save_dir   SAVE_DIR

default annot_file =./nrPDB-GO_2019.06.18_annot.tsv proposed by [1].

References

[1] Gligorijević, V., Renfrew, P.D., Kosciolek, T. et al. Structure-based protein function prediction using graph convolutional networks. Nat Commun 12, 3168 (2021).