Answers are grounded in this paper and cite their sources.
Evaluates structural risk in SARS-CoV-2 spike protein, suggesting a novel quantification approach.
Genomic surveillance pipelines typically classify SARS-CoV-2 variants against curated lineage dictionaries, creating a structural "blind window" between a variant's public sequence deposition and formal risk characterization. Here we present a zero-shot structural risk quantification framework that applies the protein language model ESM-2, without fine-tuning on epidemiological outcome labels, to estimate destabilization at functionally weighted positions of the SARS-CoV-2 spike protein, yielding a composite Aggression Score. We retrospectively evaluated this framework against four GenBank isolates spanning distinct evolutionary periods and lineage-dictionary coverage: Delta, Omicron BA.1, BF.11.3, and a 2026 isolate externally designated XFG.5.1.7. Two of these isolates lack dedicated entries in the reference signature dictionary, yet both were correctly flagged with elevated Aggression Scores. These results indicate that structural risk quantification via masked-token log-likelihood ratios can operate independently of lineage-dictionary coverage. We interpret these findings as retrospective consistency rather than prospective predictive validation.
No takes yet. Share an insight, caveat, or question.
Richard Alexander Armuelles Zeballos (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: