RINet Build 013 — Scientific Transparency Studio
Release date: 2026-08-24
Version: 1.3.0-build013
What changed
Build 013 changes RINet from a polished residue recommender into a scientific hypothesis and experiment-design system. The primary result is now a traceable argument: why a residue ranks above alternatives, which lower-signal residue is a fair comparison, what substitution probes, which integrity gates must pass, what result would support the interpretation, and what result would refute it.
The public demo and installed app now expose the graph definition, contact cutoff, exact score equation, normalized feature values, numerical contributions, deterministic tie-breaks, author residue identifiers, coordinate gaps, ligand and metal contacts, probable disulfides, data-quality fields, complete top-10 ranking, matched-control quality, mutation chemistry, benchmark rank/percentile, parameter sensitivity, sequence context, and scientific limitations.
Added for structural biologists
- explicit “what this adds beyond a structure viewer” positioning;
- scientist-level explanations of degree, closeness, betweenness, packing, long-range, and cross-chain evidence;
- linked 3D structure, author-numbered sequence, ranked table, and residue-network projection;
- hypothesis lenses for ligand sites, interfaces, allostery, stability/engineering, and antibody/CDR-aware work;
- direct metal-coordination logic distinct from generic ligand proximity;
- coordinate-derived disulfide warnings without a cysteine ranking bonus;
- residue-specific mutation chemistry, alternatives, expected contrasts, risks, and interpretation boundaries;
- matched controls, integrity gates, competing explanations, and falsification criteria;
- known-site recovery as rank/percentile, with labels applied after ranking;
- cutoff/ranking sensitivity and multi-state comparison;
- PDB and mmCIF parsing, including large/new-entry mmCIF fallback;
- chain, priority, charge, hydrophobicity, and B-factor/pLDDT coloring;
- black/white backgrounds, rotation off by default, and explicit reset/start/stop controls;
- honest not-calculated states for missing conservation, dynamics, free-energy, mutational, and systematic benchmark evidence;
- removal of generative-model prose from the scientific result.
Verification
- 72/72 automated tests passed.
- Fresh package build and distribution metadata checks passed.
- The locally installed app passed its doctor and 28-part scientific self-test.
- Browser QA passed for PDB and mmCIF uploads, 4HHB direct heme/iron coordination, 7YUE antibody/CDR-like navigation, and the public-demo workflows for 4HHB, 5CFO, 5DTL, 7YUE, and the 9XU3 PDB-to-mmCIF fallback.
- The final antibody workflow produced zero browser-console errors.
Scientific boundary
RINet is a hypothesis-triage and experiment-design layer, not a functional oracle. Static residue graphs do not replace evolutionary analysis, all-atom energetics, molecular dynamics, free-energy calculations, formal antibody numbering, mutational scanning, multiple experimentally resolved states, or biological validation. Build 013 makes those boundaries visible so absence of evidence cannot be mistaken for computed evidence.