Overview
- V2P is a multi-task model that outputs a pathogenicity score plus 23 phenotype-specific risk scores drawn from first-level Human Phenotype Ontology categories.
- The system was trained on large datasets of harmful and benign variants with disease annotations to learn links between genetic changes and clinical outcomes.
- In evaluations using de-identified patient and simulated data, the tool frequently placed the true causal mutation within the top 10 candidates.
- Benchmarking reported by the authors showed V2P outperforming other methods across phenotypes, including average AP score gains of about 0.16 over the next best per phenotype and 0.38 over the median across datasets.
- The current release assigns risks to broad disease categories, and the team plans refinements for more specific predictions and integration with additional data to support drug discovery.