A fine-tuned genomic language model captures nucleotide-level information overlooked by missense variant impact predictors.

Fine-tuned genomic language model capturing nucleotide-level information overlooked by standard missense variant impact predictors

This bioRxiv preprint presents a fine-tuned genomic language model (LLM) that captures nucleotide-level information overlooked by standard missense variant impact predictors. Existing predictors focus on protein-level consequences and share overlapping annotation priors, creating blind spots — particularly for variants acting through nucleotide sequence context (splicing, regulation). The model significantly improves pathogenicity prediction on independent benchmarks.

Synthesis written by Geno'X. For the full original abstract, please refer to the source publication.

Missense pathogenicity predictors (CADD, REVEL, AlphaMissense) are standard in genomic diagnostics but share common biases. A genomic LLM capturing complementary nucleotide-level information is a genuine advance — pending validation on independent diagnostic cohorts after peer review.

Clinical impact: 3/3 · Evidence strength: 3/3 · Novelty: 2/2 · Sample size: 1/1 · Publication status: 0/1 → Total: 9/10