DNA decoder — visual

generated 2026-07-13 · dna_decode.viz.hub · semantic zoom: genome → network → protein

What this is — and is not. A legibility layer over the deterministic decoder + its findings. Every view is READ-ONLY over committed artifacts; the frozen decoder surface is untouched. The honesty rails carry into the pixels: the network is associational, not causal (the clade de-confound is drawn as solid vs. dashed); the protein heatmap is a zero-shot molecular rank, not a resistance call (catalog markers live on a separate axis); a phenotype claim appears only behind a validated-determinant wall. Nothing here is a learned phenotype predictor.

1 Whole genome — which parts do what

point at one genome -> an evidence-tiered map of every feature (circular ring / linear browser)

scale 1 · genome

Circular genome ring — GCA 002180195.1 2026-07-13

Circos-style ring; features by strand band, determinants called out

phenotype claim only on determinant-phenotype features

2 The network — which parts relate to which

associational co-occurrence of AMR determinants, with the leave-one-clade-out de-confound drawn in

scale 2 · network

Co-occurrence network — ecoli 2026-07-13

force-directed AMR-determinant co-occurrence graph

associational, NOT causal; dashed = lineage-mediated (possible clonal structure)

scale 2 · network

Co-occurrence network — klebsiella 2026-07-13

force-directed AMR-determinant co-occurrence graph

associational, NOT causal; dashed = lineage-mediated (possible clonal structure)

3 The protein — what a specific edit does

a position x amino-acid ESM molecular-damage heatmap for one protein (zero-shot rank, not a phenotype)

scale 3 · protein

Protein damage heatmap — hiv rt 2026-07-13

position x amino-acid ESM molecular-damage heatmap

zero-shot molecular rank, NOT a resistance call; catalog markers on a separate axis