Missing covalent-bond records in the Protein Data Bank follow the ligand code: an archive-wide count, a same-day leritrelvir pair, and carry-forward across depositions

Protein–ligand covalent bonds reach training and benchmark sets through one route: rows of type covale in the entry's _struct_conn table. A bond that is present in the coordinates but absent from that table is treated as non-covalent downstream. Header files of all 199,326 PDB entries that contain a non-polymer entity hold 43,488 bonds under 2.0 Å between a Cys SG, Ser OG, Lys NZ or His NE2/ND1 atom and a ligand atom. Of these, 5,594, in 1,057 entries, have no covale row and are listed only in the same file's close-contact table, _pdbx_validate_close_contact. A coordinate census of all 1,957 SARS-CoV-2 Mpro entries confirms that the close-contact table finds these bonds: 72 of 1,109 Cys145 bonds under 2.0 Å are unrecorded, all 72 appear in the close-contact table, and none of the 1,037 recorded bonds does. Fifty-six of the 72 are in entries processed at PDBj, all from depositions made between December 2020 and February 2023. Nirmatrelvir's Cys145 bond is recorded 56 of 56 times at the other three sites and 6 of 24 times at PDBj. The ligand code is a second factor, separate from the processing site. 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. The geometry does not differ between the two groups: SG–C7 is 1.849–1.920 Å against 1.815–1.932 Å (Mann–Whitney p = 0.82). Across the archive, a code whose first bonded use went unrecorded tends to stay unrecorded. In a strict set of 5,355 (code, entry) units, the adjusted odds ratio for a later miss is 14.2 (95% CI 4.3–47). It is 8.5 (4.3–16.5), unadjusted, for reuse by other laboratories, and 12.5 (3.6–44) in the joint model with PDBj excluded. When a laboratory reused the code under which it had missed a bond, the later unit was missed 22 of 49 times; when it filed the same compound under a different code, 0 of 5 times. All of this rests on 21 first-missed codes. Three explanations were tested and none is supported. Codes first defined on a non-bonded model do not carry misses forward once commodity ligands are removed (OR 1.47, 0.67–3.21). Annotation does not copy a link template forward: the leaving-atom flag follows the dictionary in 1,481 of 1,481 and 3,559 of 3,560 rows, and refinement program and era show no effect. No feature of the dictionary entry marks the affected codes. Today's PCM annotation does separate them, but 3,454 of its modification dates fall on a single bulk update on 2024-09-27, years after most of the bonds in question were annotated. PLINDER, LP-PDBBind, the DiffDock time split and Runs N' Poses all carry these bonds as non-covalent or unlabelled. Removing the affected entries changes no published headline score by more than 0.33 percentage points. The defect is in the labels, not the leaderboards. A check that reads only the header file lists every case of this kind. This record holds the note (PDF and DOCX) and a reproduction bundle: the archive-wide bond table built from header files of 199,326 entries, the SARS-CoV-2 Mpro census tables, the leritrelvir tables, the code-history and dataset-label tables, the analysis scripts as run, and verify_numbers.py, which recomputes every number quoted in the note from the bundled tables (116/116 checks). It extends the 24,144-entry cysteine scan (doi:10.5281/zenodo.24223030). Not peer reviewed.

Authors

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-16
DOI
https://doi.org/10.5281/zenodo.24285353
Primary Topic
Enzyme Structure and Function
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Missing covalent-bond records in the Protein Data Bank follow the ligand code: an archive-wide count, a same-day leritrelvir pair, and carry-forward across depositions

Michael T. Varman
2 links
Zenodo (CERN European Organization for Nuclear Research)
Enzyme Structure and Function
article

Missing covalent-bond records in the Protein Data Bank follow the ligand code: an archive-wide count, a same-day leritrelvir pair, and carry-forward across depositions

Michael T. Varman
article en

Abstract

Protein–ligand covalent bonds reach training and benchmark sets through one route: rows of type covale in the entry's _struct_conn table. A bond that is present in the coordinates but absent from that table is treated as non-covalent downstream. Header files of all 199,326 PDB entries that contain a non-polymer entity hold 43,488 bonds under 2.0 Å between a Cys SG, Ser OG, Lys NZ or His NE2/ND1 atom and a ligand atom. Of these, 5,594, in 1,057 entries, have no covale row and are listed only in the same file's close-contact table, _pdbx_validate_close_contact. A coordinate census of all 1,957 SARS-CoV-2 Mpro entries confirms that the close-contact table finds these bonds: 72 of 1,109 Cys145 bonds under 2.0 Å are unrecorded, all 72 appear in the close-contact table, and none of the 1,037 recorded bonds does. Fifty-six of the 72 are in entries processed at PDBj, all from depositions made between December 2020 and February 2023. Nirmatrelvir's Cys145 bond is recorded 56 of 56 times at the other three sites and 6 of 24 times at PDBj. The ligand code is a second factor, separate from the processing site. 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. The geometry does not differ between the two groups: SG–C7 is 1.849–1.920 Å against 1.815–1.932 Å (Mann–Whitney p = 0.82). Across the archive, a code whose first bonded use went unrecorded tends to stay unrecorded. In a strict set of 5,355 (code, entry) units, the adjusted odds ratio for a later miss is 14.2 (95% CI 4.3–47). It is 8.5 (4.3–16.5), unadjusted, for reuse by other laboratories, and 12.5 (3.6–44) in the joint model with PDBj excluded. When a laboratory reused the code under which it had missed a bond, the later unit was missed 22 of 49 times; when it filed the same compound under a different code, 0 of 5 times. All of this rests on 21 first-missed codes. Three explanations were tested and none is supported. Codes first defined on a non-bonded model do not carry misses forward once commodity ligands are removed (OR 1.47, 0.67–3.21). Annotation does not copy a link template forward: the leaving-atom flag follows the dictionary in 1,481 of 1,481 and 3,559 of 3,560 rows, and refinement program and era show no effect. No feature of the dictionary entry marks the affected codes. Today's PCM annotation does separate them, but 3,454 of its modification dates fall on a single bulk update on 2024-09-27, years after most of the bonds in question were annotated. PLINDER, LP-PDBBind, the DiffDock time split and Runs N' Poses all carry these bonds as non-covalent or unlabelled. Removing the affected entries changes no published headline score by more than 0.33 percentage points. The defect is in the labels, not the leaderboards. A check that reads only the header file lists every case of this kind. This record holds the note (PDF and DOCX) and a reproduction bundle: the archive-wide bond table built from header files of 199,326 entries, the SARS-CoV-2 Mpro census tables, the leritrelvir tables, the code-history and dataset-label tables, the analysis scripts as run, and verify_numbers.py, which recomputes every number quoted in the note from the bundled tables (116/116 checks). It extends the 24,144-entry cysteine scan (doi:10.5281/zenodo.24223030). Not peer reviewed.

Zenodo (CERN European Organization for Nuclear Research)
Enzyme Structure and Function
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon