Training set for RosettaFold/MPNN etc.

Each PDB entry is represented as a collection of .pt files:
    PDBID_CHAINID.pt - contains CHAINID chain from PDBID
    PDBID.pt         - metadata and information on biological assemblies

PDBID_CHAINID.pt has the following fields:
    seq  - amino acid sequence (string)
    xyz  - atomic coordinates [L,14,3]
    mask - boolean mask [L,14]
    bfac - temperature factors [L,14]
    occ  - occupancy [L,14] (is 1 for most atoms, <1 if alternative conformations are present)

PDBID.pt:
    method        - experimental method (str)
    date          - deposition date (str)
    resolution    - resolution (float)
    chains        - list of CHAINIDs (there is a corresponding PDBID_CHAINID.pt file for each of these)
    tm            - pairwise similarity between chains (TM-score,seq.id.,rmsd from TM-align) [num_chains,num_chains,3]
    asmb_ids      - biounit IDs as in the PDB (list of str)
    asmb_details  - how the assembly was identified: author, or software, or smth else (list of str)
    asmb_method   - PISA or smth else (list of str)

    asmb_chains    - list of chains which each biounit is composed of (list of str, each str contains comma separated CHAINIDs)
    asmb_xformIDX  - (one per biounit) xforms to be applied to chains from asmb_chains[IDX], [n,4,4]
                     [n,:3,:3] - rotation matrices
                     [n,3,:3] - translation vectors

list.csv:
   CHAINID    - chain label, PDBID_CHAINID
   DEPOSITION - deposition date
   RESOLUTION - structure resolution
   HASH       - unique 6-digit hash for the sequence
   CLUSTER    - sequence cluster the chain belongs to (clusters were generated at seqID=30%)
   SEQUENCE   - reference amino acid sequence


valid_clusters.txt - clusters used for validation

test_clusters.txt - clusters used for testing