Bonds inside the processing distance window that the Protein Data Bank still leaves unrecorded, and how much of them the ligand code's history predicts
Year: 2026
Type: article
Abstract: My distance-table note (Varman 2026e) showed that whether the Protein Data Bank records a protein–ligand covalent bond follows a distance table in its processing software. MAXIT's _allowed_bonds table gives one link window per element pair, for example C–S 1.55–2.00, C–O 1.10–1.55, C–N 1.20–1.60, S–S 1.85–2.20 and O–P 1.40–1.75 Å. Bonds outside a window are listed as close contacts, and in that note's X-ray set 4.5% of residue-pair contacts inside a window (511 of 11,325) are unrecorded with no explanation from the table. This addendum asks which in-window bonds still go unrecorded, and whether the history of the ligand's code predicts them. In the header files of all 199,326 PDB entries with a non-polymer entity, 40,899 contacts between a Cys SG, Ser OG, Lys NZ or His NE2/ND1 atom and a ligand residue fall inside one of the ten windows these atoms can reach. Of these, 2,971 (7.3%) have no covale row in _struct_conn. Bilins account for 1,655 of them; without bilins the rate is 4.6% (1,316 of 28,851 contacts). Entries processed at PDBj hold 1,965 of the 2,971. All 72 unrecorded Cys145 bonds in a complete census of SARS-CoV-2 Mpro lie inside the C–S window, at 1.696–1.99 Å. The ligand code is one predictor. On 11 February 2026 one laboratory deposited 16 leritrelvir–Mpro entries, all processed at PDBc. The 11 entries filed under CCD code 7ON have the bond recorded in 22 of 22 copies; the 5 filed under A1EN0 have it in 0 of 10. Both groups lie inside the C–S window (SG–C7 1.849–1.920 Å against 1.815–1.932 Å), so A1EN0 belongs to the in-window misses that the distance table does not explain. Across the archive, restricted to in-window contacts, a code whose first bonded use went unrecorded tends to stay unrecorded. In a strict set of 5,611 (code, entry) units with 299 misses, the adjusted odds ratio for a later miss is 16.7 (95% CI 5.1–54.3). For reuse by other laboratories the Fisher odds ratio is 10.3 (5.3–20.1), and with PDBj excluded the joint-model odds ratio is 15.2 (3.8–60.3). A laboratory that filed the same compound under a different code after a miss recorded the bond 5 of 5 times. The effect is real but narrow. It rests on 19 first-missed codes with later uses, and two of them, HUR and 80I, both at PDBj, supply 15 of the 25 later misses. Without them the joint-model odds ratio is 6.7 (1.6–28.5). Code history accounts for a small share of in-window misses. In the strict set, 25 of 299 missed units (8.4%) are later uses of a first-missed code, about 21 more than expected. Across all in-window units the figure is 61 of 552 (11.1%). Most in-window misses in the strict set fall on a code's first deposition day (197 of 299) or under codes whose first use was recorded (77). Explanations tested in v1.0 and not supported are summarised here without being rerun. A check that reads only the header file can list these cases for curation.
Author: Varman, Michael T.
Language: English
Subfield: Structural Biology
Domain: Life Sciences
Keywords: bilins, SARS-CoV-2 Mpro
Open Access status: green
OpenAlex ID: W7234430119
