No BLAST, New Problems! Auditing Homology-Free Protein-Function Prediction

Deciding whether a mutation changes protein function drives variant interpretation and protein design, so a method that answers without a database match must carry its own evidence. Sequence-homology searches provide both predictions and an external biological reference, but $30$--$50\%$ of genes lack annotated homologs.\textsuperscript{1} Protein language models offer a database-free alternative by treating movement in embedding space as evidence of functional change. Removing the database match, however, also removes the reference that would expose a misleading score. \textbf{We test whether a homology-free claim survives an adversarial audit of its data, generator, generalization, attribution, and metric.} PRISM\textsuperscript{5} applies 78 COSMIC mutational signatures\textsuperscript{2} to coding sequences, embeds wild-type and mutant proteins with ESM-2,\textsuperscript{3} and measures movement toward Pfam-domain centroids.\textsuperscript{4} Its strongest reported result links SBS17a to the PF17041 domain family across two structurally distinct protein clans. We reconstructed the analysis from archived outputs and designed a matched control for each interpretive step.