Protein structure predictions for Pfam families
===============================================

This directory provides structural models and contact maps 
for 6,370 Pfam families (Pfam 33.1).

Models were generated by the Baker group (https://www.bakerlab.org/) 
using trRosetta (1) from multiple sequence alignments.

Overview
--------

Download the archive and verify its integrity:

    $ wget http://ftp.ebi.ac.uk/pub/databases/Pfam/baker/pfam33.1_baker.tar.gz
    $ wget http://ftp.ebi.ac.uk/pub/databases/Pfam/baker/pfam33.1_baker.tar.gz.md5
    $ md5sum -c pfam33.1_baker.tar.gz.md5

Untar the archive:

    $ tar -xzf pfam33.1_baker.tar.gz
    $ ls -l 
        drwxr-xr-x 5 interpro interpro        4096 Feb 25 22:49 pfam33.1_baker
        -rw-r--r-- 1 interpro interpro 86271431608 Feb 25 22:49 pfam33.1_baker.tar.gz
        -rw-r--r-- 1 interpro interpro          57 Feb 25 22:49 pfam33.1_baker.tar.gz.md5

The pfam33.1_baker directory has the following structure:

    .
    |-- a3m
    |   |-- PF00242.a3m
    |   |-- PF00257.a3m
    |   |-- ...
    |   `-- PF19222.a3m
    |-- dist
    |   |-- PF00242.npz
    |   |-- PF00257.npz
    |   |-- ...
    |   `-- PF19222.npz
    |-- lddt
    |   |-- PF00242.npz
    |   |-- PF00257.npz
    |   |-- ...
    |   `-- PF19222.npz
    `-- pdb
        |-- PF00242.pdb
        |-- PF00257.pdb
        |-- ...
        `-- PF19222.pdb


    * a3m: contains the UniProt alignments for Pfam families, in the A3M file format.
    
    * dist: contains predicted protein inter-residue distances and orientation, 
            in the NumPy NPZ file format.

    * lddt: contains predicted residue-wise local Distance Difference Test (lDDT) scores (2), 
            in the NumPy NPZ file format.            
    
    * pdb: directory of predicted structures, in the PDB file format.

Contact maps
------------

Contact maps can be retrieved from the files in the /dist/ directory, 
using the Python code below (requires the NumPy package).

    $ python
    Python 3.6.9 (default, Oct  8 2020, 12:12:24) 
    [GCC 8.4.0] on linux
    Type "help", "copyright", "credits" or "license" for more information.
    >>> import numpy as np
    >>> data = np.load("pfam33.1_baker/dist/PF00242.npz")
    >>> dist = data["dist"]
    >>> matrix = np.sum(dist[:,:,1:13], axis=-1)

The dist array represents the residue-reidue distance and is of shape [L, L, 37]
where L represents the size of the first sequence of the alignment.
The third dimension represents probabilities for the Cb-Cb distance 
to be within the ranges: 0-2.5A, 2.5-3A, ..., 19.5-20A, >20A.

matrix is the contact map with probabilities of Cb-Cb distances <8A.

lDDT scores
-----------

The local Distance Difference Test (lDDT) scores can be retrieved from the files
in the /lddt/ directory.

    $ python
    Python 3.6.9 (default, Oct  8 2020, 12:12:24) 
    [GCC 8.4.0] on linux
    Type "help", "copyright", "credits" or "license" for more information.
    >>> import numpy as np
    >>> data = np.load("pfam33.1_baker/lddt/PF00242.npz")
    >>> lddt = data["lddt"]

lddt is an array of the size of the sequence.

References
----------

(1) Yang J, et al. Improved protein structure prediction using predicted interresidue orientations. Proc Natl Acad Sci U S A. 2020
(2) Mariani V, et al. lDDT: a local superposition-free score for comparing protein structures and models using distance difference tests. Bioinformatics. 2013