MgKdb: A dataset of curated magnesium and potassium ions annotated on cyo-em structure

MgKdb — Mg²⁺ / K⁺ ion database Per-site records of every Mg²⁺ and K⁺ ion annotated in a set of 1,052 cryo-EM structures, classified by Cat_Wiz, together with the atomic model and density box around each validated site and the data supporting the technical validation. Every structure was retrieved from the PDB with its corresponding EMDB map and processed with Cat_Wiz, which diagnoses each ion from its coordination stereochemistry and sorts it into two classes: validated — the site meets the criteria for a reliable assignment. check — one or more criteria were not met. This is a flag for manual inspection, not a rejection: the class contains both genuine errors and sites that are merely incompletely modelled (typically missing waters). validated check total Mg²⁺ 12,302 125,508 137,810 K⁺ 2,852 10,628 13,480 Total 15,154 136,136 151,290 Classification criteria, in the distance bins Cat_Wiz uses (< 2.3 / 2.3–2.6 / 2.6–3.2 Å): Mg²⁺ validated — exactly six ligands below 2.3 Å, i.e. a complete octahedral inner shell. K⁺ validated — every contact inside the 2.6–3.2 Å window, none shorter, and Cat_Wiz's own High certainty marker (≥ 5 contacts). The window criterion alone proved too permissive: three independent curators confirmed only 54% of it, against 87% once the marker is applied. Folders path contents mg/ Mg²⁺ site tables, plus one 5 Å box per validated site mg/atoms/ 12,302 .cif — atomic model around each validated Mg²⁺ mg/maps/ 12,302 .map — density around each validated Mg²⁺ k/ K⁺ site tables, plus one 5 Å box per validated site k/atoms/ 2,852 .cif k/maps/ 2,852 .map other_ions/ every other monatomic ion found in the same structures, with boxes other_ions/atoms/ 4,047 .cif other_ions/maps/ 4,047 .map technical_validation/CMM/ CheckMyMetal reports and the per-ion comparison technical_validation/manual_curation/ blind manual review by three curators technical_validation/mic/ size comparison against the MIC dataset other_ions/ is provided for completeness only. Cat_Wiz diagnoses Mg²⁺ and K⁺ exclusively, so none of these sites carries a validated/check label and none has been validated here. Site identifiers Every site has an id of the form <PDB>_<chain><U|L>_<resno>_<ION> e.g. 7K00_aL_6083_MG U/L records the case of the chain identifier so that chains differing only in case (A vs a) do not collide on case-insensitive filesystems. The same id is the file name of the corresponding .cif and .map. Boxes Each box is a 10 Å cube centred on the ion (5 Å in every direction). atoms/*.cif holds only atoms whose centre falls inside the box, so residues are truncated at the box edge. mmCIF rather than PDB because ribosomal auth chain identifiers are multi-character (AN1, 16S1) and do not fit the fixed-column PDB format. maps/*.map is a CCP4 sub-map at the native voxel spacing of the source EMDB map — not resampled. It references the original unit cell with the crop's integer grid offset stored in NXSTART (the crystallographic convention), so the box overlays the model correctly in ChimeraX, Coot, PyMOL and 3Dmol. A standalone small-cell map with an MRC ORIGIN header does not work: ChimeraX treats the small P1 cell as crystallographic, ignores ORIGIN, and places the box at the origin. File formats All tables are tab-separated with a header row. Shared columns These appear in mg_validated.tsv, mg_check.tsv, k_validated.tsv and k_check.tsv, and are described once here. column meaning id site identifier (see above) pdb_id, emdb_id source structure and map accessions chain, resno chain identifier and residue number of the ion resolution_bin high (<2 A), medium (2-2.5 A) or low (2.5-3 A) structure_type annotation from the PDB entry, e.g. ribosome cn coordination number as reported by Cat_Wiz rmsd_ion Cat_Wiz RMSD figure for the site binding_site_type ligand composition, e.g. cis-2Oph.4Ow, 6Ow quality Cat_Wiz confidence marker (see per-file notes) n_water_ligands number of coordinating waters na_prot environment: NA, Prot. or NA/Prot. x, y, z ion position, in the frame of the deposited model Ligand descriptions are slash-separated lists aligned position by position: the n-th entry of ligand_atoms*, ligand_residues* and ligand_chains* describes the same contact, in the same order as the distances. w denotes water. mg/mg_validated.tsv — 12,302 rows, 29 columns A validated Mg²⁺ has exactly six ligands below 2.3 Å by definition, so the 2.3–2.6 and 2.6–3.2 bins are always empty and are not carried over; cn is always 6. column meaning distances the six contact distances, slash-separated dist_min, dist_max, dist_mean derived from distances rmsd_ligands per-ligand RMSD, same order as distances dist_oct deviation from ideal octahedral geometry clash, has_clash contact in the 3.2–3.6 Å range, if any quality High, Fair or Low ligand_atoms, ligand_residues, ligand_chains the six coordinating atoms mg_mg_pairs, mg_k_pairs neighbouring ions, as distance/chain:resno k/k_validated.tsv — 2,852 rows, 28 columns Same shape as mg_validated.tsv, with three differences: contacts lie in the 2.6–3.2 Å window rather than below 2.3 Å, so n_contacts gives the coordination count; there is no dist_oct, since K⁺ has no octahedral requirement; and contact_type replaces clash, describing the contact topology (/Oph-bridge, /G-U-cWW, /Prot.). quality is always High certainty — that is the criterion for the class. Neighbouring ions are in k_k_pairs and k_mg_pairs. mg/mg_check.tsv — 125,508 rows, 42 columns · k/k_check.tsv — 10,628 rows, 39 columns Check sites failed at least one criterion, so contacts may fall in any of the three bins and every measurement is split accordingly. Where the validated tables have one column, these have three, suffixed _lt23, _23_26, _26_32: n_* (counts, also summarised in n_total), dist_*, rmsd_*, ligand_atoms_*, ligand_residues_*, ligand_chains_*. cn is given as a/b/c over the same three bins. column meaning flag_reason why Cat_Wiz raised the site — the most informative column quality mostly empty for check sites flag_reason values are Poorly modelled coordination shell, Check if M+ ion, Possible hexahydrated Mg2+ ion, Incomplete coordination shell, Check/Complete hydration shell and Check/Remove water for Mg²⁺; Check coordination and Check K+ for K⁺. In k_check.tsv, quality also records provenance: Lower certainty marks the 7,250 sites demoted from the validated block, while sites Cat_Wiz flagged directly carry an empty value or To distant. Mg²⁺-specific columns (dist_oct, clash, has_clash, clash_info) are absent from k_check.tsv, which carries contact_type instead. structures_pdbid.tsv — 1,052 rows One row per analysed structure: pdb_id, emdb_id, resolution_bin, structure_type, total_atoms, then ion counts per class (mg_validated, mg_check, mg_total, k_validated, k_check, k_total) and the flags has_mg, has_k. Column sums reproduce the table above. other_ions/other_ions.tsv — 4,047 rows id, pdb_id, chain, resno, ion (residue name), element, x, y, z, and has_files (whether the box was extracted). ion element sites ion element sites ZN zinc 3,318 CU1 copper 5 NA sodium 634 FE iron 4 CL chloride 72 CA calcium 3 MN manganese 8 PT platinum 2 BR bromide 1 Only identified species are included. UNX — the PDB code for an unknown atom or ion, carrying element X — is not an ion and is excluded, along with the other unknown-atom placeholders (UNK, UNL). Technical validation technical_validation/manual_curation/ reviewer_1.tsv, reviewer_2.tsv, reviewer_3.tsv — 400 rows each. Three curators independently reviewed the same 400 ions (300 Mg²⁺, 100 K⁺) blind, in a 3D viewer showing model and map with an adjustable contour level, without seeing the Cat_Wiz label. column meaning reviewer anonymised curator identifier id, ion the site judged decision the curator's answer: validated or check catwiz_class the Cat_Wiz class agree whether the two match contour contour level, in σ, at the moment of the decision timestamp when the decision was recorded Purity of the validated class — the fraction each curator confirmed: Mg²⁺ (n = 150) K⁺ (n = 23) reviewer_1 98.7% 95.7% reviewer_2 99.3% 95.7% reviewer_3 100.0% 91.3% all three 98.7% 87.0% technical_validation/CMM/ json/ holds the raw CheckMyMetal report for each of 100 sampled structures. cmm_summary.tsv — 19,366 rows — pairs each CMM site with its Cat_Wiz class: column meaning pdb_id, id, ion the site catwiz_class validated or check cmm_proposed_metals CMM's ranked proposals, Metal(score), comma-separated cmm_suggested CMM scores the deposited element above zero cmm_main_suggested the deposited element is top-ranked cmm_suggested is not "the element appears in the list": CMM prints every metal it considered and scores the rejected ones 0, so presence alone is 100% by construction. For cmm_main_suggested, metals sharing the maximum score all count as top-ranked, but only when that maximum is above zero — a site where every metal scores 0 has no proposal, and nothing is credited. validated check Mg²⁺ (n) 2,923 14,312 CMM main suggested Mg 82% 10% CMM suggested Mg 100% 55% K⁺ (n) 919 1,212 CMM main suggested K 98% 85% CMM suggested K 100% 97% technical_validation/mic/ Size comparison against MIC (Metric Ion Classification, Nat Commun 2025, doi:10.1038/s41467-025-61315-x). mic_comparison.tsv holds the counts, mic_dataset_composition.tsv the MIC sites per ion class, mic_scope.tsv how the two resources differ, and MIC_SupplementaryData1.xlsx the source data from Zenodo (10.5281/zenodo.15596177) — the paper publishes only the ranking of classes, not the counts, so the figures are recomputed from this file. This is a comparison of scale and coverage, not of predictive performance. The two resources sample disjoint regions of structural space and share no structures: MIC is protein / X-ray / high resolution and classifies from coordinates alone, whereas MgKdb is RNA / cryo-EM and uses the density map. Notes and caveats check is not a rejection. Manual curation confirmed a substantial share of check sites as sound, so global agreement between a curator and Cat_Wiz is not a meaningful quality metric. The purity of the validated class is. Coordination counts are censored by unmodelled waters. In cryo-EM, waters are frequently absent from the deposited model, so the number of contacts observed is a lower bound on the true coordination number. This is why the K⁺ validated fraction falls steeply with resolution (55.0% → 22.0% → 7.1% across the three bins) and why applying a contact-count criterion implicitly biases the K⁺ set towards high-resolution structures. Deposited models contain artefacts. Some entries carry duplicated ions at impossible separations — 9B24, for example, has 21 Mg–Mg pairs closer than 2.6 Å (shortest 1.12 Å), all at full occupancy with no altloc. Of 1,059 target structures, 1,052 produced Cat_Wiz output; the remaining 7 are not represented here.