Panel scope and source family assignments
Start with three residue-level cases: R252H: surface availability and uptake, S267F: opposite effects on two substrates, and E257A: reference sodium-site context and reduced uptake. These are bounded explanations, not complete transport cycles.
Explanations we can test
“Supported” is an evidence assessment, not expert sign-off or proof of a unique mechanism. “Challenged” rejects a particular explanation, not the experimental observation.
1. Observed coordinates → computed contact candidates
Identical Changed Unresolved · Cells are sequence-derived candidates, not observed target binding sites. Select a cell for its evidence. A backbone contact does not make a substitution harmless.
Exact contact atoms and source context
2. Apo predictions are a separate comparison
Sequence pairing comes first. The global CA fit uses all available consensus pairs with target pLDDT ≥70, without outlier trimming. No experimental ion is transferred into a predicted model.
Large residuals can reflect correspondence error, conformational differences, or model error. Even NTCP's own apo prediction need not match its substrate-bound conformation. pLDDT is local confidence, not certainty in the alignment between proteins.
3. What did the experiments measure?
Binding, uptake, inhibition, and infection are separate outcomes. Conflicting studies retain their own contexts.
Substitutions use reference-sequence coordinates, not a claim that the tested construct is wild type. A detected reduction in uptake is not a statement that all uptake is absent.
4. Small, source-backed mechanisms
Physical interaction edges describe binding. Substitution edges describe measured consequences; the R252H localization-to-uptake edge is explicitly an author interpretation. Missing binding, folding, or gating steps are not filled in from contact maps.
5. Residue chemistry, local context and measured outcomes
Reference chemical properties are not observations at a protein site. The catalog review is machine-assisted source-transcription checking, not independent human approval. Published simulations, reference-coordinate neighborhoods and assays remain separate; unresolved mediation links are not drawn as causal edges.
Reuse an explanation without losing its limits
Use a stable assertion ID together with the bundle version and SHA-256 in the export manifest. Every claim has an evidence origin, review status, source references, and limitations. A read-only check against the commit-pinned SLC10A4 gene review matched its seven retained residue identities and correspondences. This does not endorse its GO annotation decisions; the upstream repository has not been modified to adopt PTM.