PLLM_PDB_Dataset

PDB 3Di Chains Dataset

This repository contains chain-level sequences and 3Di tokens derived from RCSB PDB structures, with per-chain polymer class labels (prot, DNA, RNA, other). Files are chunked/merged from 1k-folders and cleaned for consistent CSV schema. Current data based on only 120K proteins.

Files

3di_chains_chaintag_filtered_prot_only.csv <- Use this one, the main file

3di_chains_chaintag_all.csv
474 MB

3di_chains_chaintag_prot_only.csv
464 MB

3di_chains_chaintag_sample_1000.csv

• 3di_chains_chaintag_filtered_prot_only.csv filtered out repeated protein and DNA RNA chain. Main file used for SFT
• _all contains all chains (including D-amino, RNA, DNA, and any others).
• _prot_only contains protein chains only (L- and D-amino acids treated as protein).
• _sample_1000 is a 1,000-row random sample for quick inspection.

All raw PDB files are in https://drive.google.com/drive/folders/1jdz5c_EoNCpqXXmDdklr1tlZtzN8b4jY?usp=sharing

Schema

All CSVs use the same columns:

• index — global row index from the source CSV (0-based)
• pdb_id — 4-character PDB code (e.g., 9B4J)
• chain_id — chain identifier (alphanumeric, may include digits/letters)
• aa_seq — amino-acid sequence (when available)
• threeDi_seq — Foldseek 3Di token sequence (Used for sft)
• combined_seq — helper concatenation (3Di/AA) used upstream
• seq_len — chain sequence length (prefer AA length; else derived)
• chunk — source folder name (e.g., 1000_1999)
• path — absolute path to the structure file used
• polymer_class — one of prot, DNA, RNA, other
  (D-amino peptides are classified as prot)

Basic Stats (from _all)

=== BASIC COUNTS === Proteins (unique pdb_id) : 111629 Chains (rows) : 418335 Avg chains per protein : 3.75

=== CHAIN-LEVEL COMPOSITION === prot: 391317 (93.54%) DNA: 18575 (4.44%) RNA: 8375 (2.00%) other: 68 (0.02%)

=== PROTEIN-LEVEL PRESENCE (non-exclusive) === Proteins with any prot: 109267 (97.88%) Proteins with any DNA : 7334 (6.57%) Proteins with any RNA : 4622 (4.14%)

=== CHAIN LENGTH (residues) === mean: 254.88 | Q25: 103 | Q50: 208 | Q75: 335

=== PROTEIN TOTAL LENGTH (sum of chains) === mean: 955.19 | Q25: 270 | Q50: 516 | Q75: 1089