Researchers at the Icahn School of Medicine at Mount Sinai have introduced a new artificial intelligence model that could reshape genetic diagnostics and precision medicine. The tool, V2P (Variant to Phenotype), goes a step beyond existing technologies by not only identifying potentially harmful genetic variants but also predicting the type of disease those variants are likely to cause.
“Our approach allows us to pinpoint the genetic changes most relevant to a patient’s condition,” explains first author David Stein, PhD. “By predicting both pathogenicity and disease category, we can improve the speed and accuracy of genetic interpretation.” The work was published in Nature Communications.
Next-generation approach
Traditional computational tools estimate whether a mutation is pathogenic, but they rarely provide insight into what condition that mutation might trigger. V2P fills this critical gap by using advanced machine learning models trained on large datasets of benign and pathogenic variants, combined with disease-specific information.
When tested on de-identified patient datasets, V2P consistently ranked the true disease-causing variant among the top candidates. A signal that it could dramatically streamline the diagnostic workflow for complex and rare disorders.
Drug discovery and precision medicine
Beyond diagnostics, V2P could become a powerful tool for identifying new therapeutic targets. By linking variants to disease mechanisms, researchers can focus on the most biologically relevant genes and pathways. “This model can help guide genetically informed drug discovery,” says co-senior author Avner Schlessinger, PhD, Director of the AI Small Molecule Drug Discovery Center at Mount Sinai. “It enables us to connect genetic disruptions with the biological systems they affect, which is especially valuable in rare and difficult-to-treat diseases.”
Although V2P currently categorizes diseases into broad groups, such as neurological disorders or cancers, the team aims to refine its granularity and integrate additional datasets to further enhance predictive power.
Personalised care
By strengthening the link between genotype and phenotype, V2P pushes the field closer to tailored care pathways, where treatments are matched to an individual’s genetic blueprint.
“V2P gives us a clearer window into how genetic changes translate into disease,” adds co-senior author Yuval Itan, PhD. “This helps prioritise which variants and pathways deserve deeper investigation and ultimately supports more personalised, mechanism-based therapies.” The study, “Expanding the utility of variant effect predictions with phenotype-specific models,” underlines the growing role of AI in accelerating diagnostics and enabling precision healthcare.
Advanced AI models
Last month, researcher Konstantina Tzavella (Vrije Universiteit Brissel), developed advanced AI models that more accurately predict how genetic mutations alter the structure and function of proteins. These are molecules essential to virtually every biological process. Her Ph.D. work, conducted within the TumorScope project and the VUB/ULB Interuniversity Institute of Brussels, demonstrates how these models can accelerate the identification of disease-causing mutations and support personalized medicine.
Central to her research are protein language models (pLMs), AI systems that learn the “grammar” of proteins by analyzing millions of amino acid sequences. These models can detect subtle patterns that reveal how mutations might change protein behavior. Tzavella’s findings show that pLMs outperform traditional tools, especially in modeling epistasis, how combinations of mutations interact in unexpected ways.
