Datasets:
SSP: protein secondary structure (TAPE training data + CASP12-16 test sets, rcsb/casp/tape versions)
1c12c1a verified | """How TAPE writes modified residues: compare TAPE's sequences with the PDB entries' canonical sequences. | |
| For each TAPE chain whose PDB entry is available locally and whose canonical sequence has the same length, list every | |
| position where TAPE's letter differs from the PDB's canonical letter, with the PDB residue name at that position | |
| (e.g. MSE = selenomethionine). Read-only. | |
| usage: python -I scripts/check_modified_residues.py | |
| CASP12: TAPE's 21 chains, entries in data/original/rcsb (downloaded by build_ssp.py) | |
| train: TAPE training chains whose entries are in data/original/netsurfp_train_pdb (200 random entries, seed 0, | |
| downloaded for the NetSurfP label check) | |
| """ | |
| import os, sys, glob, json, collections | |
| sys.path.insert(0, "scripts") | |
| import gemmi | |
| import ssp_lib as L | |
| TAPE_DIR = "data/original/tape/secondary_structure/secondary_structure" | |
| def compare(rows, cif_dir): | |
| n = same = 0 | |
| diff, skipped = collections.Counter(), [] | |
| for r in rows: | |
| code, chain = r["id"].split("-", 1) | |
| try: | |
| st, can, _ = L.rcsb_chain(f"{cif_dir}/{code}.cif.gz", chain) | |
| except Exception as e: # chain renamed or entity layout changed in today's entry | |
| skipped.append((r["id"], f"{type(e).__name__}")); continue | |
| mon = [gemmi.Entity.first_mon(m) for m in st.get_entity_of(st[0][chain].get_polymer()).full_sequence] | |
| s = r["primary"] | |
| if len(s) != len(can): | |
| skipped.append((r["id"], f"length {len(s)} vs {len(can)}")); continue | |
| n += 1; same += s == can | |
| for a, b, m in zip(s, can, mon): | |
| if a != b: | |
| diff[(a, b, m)] += 1 | |
| t = gemmi.find_tabulated_residue(m) | |
| if t is None or not t.is_standard(): # every modified residue, whatever TAPE wrote there | |
| modified[(m, "TAPE X" if a == "X" else f"TAPE {a}")] += 1 | |
| return n, same, diff, skipped | |
| modified = collections.Counter() | |
| for name, fn, cif_dir in [("CASP12", "secondary_structure_casp12.json", "data/original/rcsb"), | |
| ("train", "secondary_structure_train.json", "data/original/netsurfp_train_pdb")]: | |
| have = {os.path.basename(f)[:4] for f in glob.glob(f"{cif_dir}/*.cif.gz")} | |
| rows = [r for r in json.load(open(f"{TAPE_DIR}/{fn}")) if r["id"][:4] in have] | |
| modified.clear() | |
| n, same, diff, skipped = compare(rows, cif_dir) | |
| print(f"== {name}: {n} chains compared ({len(skipped)} skipped: {skipped}); identical to the PDB canonical sequence: {same}") | |
| print(f" positions where TAPE differs: {sum(diff.values())}") | |
| for (a, b, m), c in diff.most_common(): | |
| print(f" TAPE {a} PDB canonical {b} PDB residue {m}: {c}") | |
| print(f" modified (non-standard) residues in these chains: {sum(modified.values())}; what TAPE wrote there: " | |
| f"{dict(collections.Counter(k[1] for k in modified.elements()))}") | |
| print(f" by residue: {dict(modified.most_common())}") | |