What this is. An associational co-occurrence network of AMR determinants within klebsiella (307 genomes). An edge means two determinants tend to appear in the same genome — NOT that one causes the other. No edge is a causal arrow. The de-confound is in the drawing. A solid node border = the association survives leave-one-clade-out CV (de-confounded). A dashed border / dashed edge = clade-concentrated: the co-occurrence may just be clonal population structure, not a real linkage. Dotted = untested. Scope + provenance. verdict PASS_LINKAGE_STRUCTURE; edges pruned to cooc ≥ 8; read-only over committed artifacts (determinant_cooccurrence_result_2026-07-11.json, crossaxis_lineage_deconfound_determinant_klebsiella_2026-07-12.json). The frozen decoder surface is untouched. caveat: Cohorts are DRUG-R/S-SELECTED (resistance-enriched) -> co-occurrence reflects the curated cohorts + selection, NOT a random population sample. Linkage is 'within these cohorts'. caveat: Within-organism de-confounds SPECIES; --dedup-profiles is only a CRUDE clonality proxy (identical determinant sets). Full Mash-clonality correction is the follow-on (needs the assemblies + Docker). caveat: AMRFinder point-mutation determinants (gyrA_S83L) are chromosomal/organism-specific; acquired genes (sul1, tet(A)) are mobile/plasmid. Linkage mixes both mechanisms.
node border (de-confound):generalizes beyond lineage (de-confounded — survives clade-grouped CV)lineage-mediated (clade-concentrated — may be clonal structure)untested (no cross-axis leave-one-clade-out entry)
fill (gene family):QRDR (cipro)beta-lactamaminoglycosidesulfa/trimethoprimtet/macrolideother