Answers are grounded in this paper and cite their sources.
This technical note examines the sensitivity of predicted protein stability estimates to sequential post-hoc structural filtering in a fixed BioEmu benchmark archive. The analysis includes 154 systems (42 wild-type and 112 mutant systems) and 112 matched mutant–wild-type comparisons. The saved ensemble contained 302,800 frames. Official geometry filtering retained 298,188 frames, followed by an additional fixed peptide-ω screen retaining 272,381 frames. Absolute changes in predicted ΔΔG of at least 0.1 kcal/mol occurred in 19 of 112 comparisons after geometry filtering and 40 of 112 comparisons after peptide-ω filtering, with 12 comparisons shared between these groups. The corresponding descriptive mean absolute errors against the benchmark experimental references were 0.702595, 0.696142, and 0.707681 kcal/mol across the three sequential ensembles. System-level foldedness, predicted ΔG, matched ΔΔG, and cross-stage comparisons reveal how changes in retained conformations can be amplified or cancelled through shared wild-type references. These results document sensitivity of a fixed estimator to the selected structural subsets; they do not demonstrate a general improvement in predictive accuracy or recovery of equilibrium conformational populations. The record contains a Technical Note, Supplementary Information, fixed-archive results, analysis and verification code, and provenance documentation. The scientific conclusions are limited to the archived inputs and specified filtering conditions. Funding: No external funding. Conflicts of interest: None declared. AI disclosure: Generative AI assisted with research planning, analysis scripting, verification support, figure preparation, and manuscript drafting. The human author reviewed the work and retains responsibility for scientific interpretation and final content. Licensing: Original author-written text and figures are CC BY 4.0; original author-owned code is MIT-licensed. Third-party code and data retain their respective upstream license conditions, as detailed in the included license and provenance files.
No takes yet. Share an insight, caveat, or question.
Sho Miyazaki (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: