Harvard researchers have developed popEVE, a powerful AI model that ranks genetic variants by disease severity, enabling faster rare-disease diagnoses and identifying more than 100 previously unknown pathogenic mutations. (Source: Image by RR)

Early Clinical Tests Show popEVE Can Shorten Diagnostic Timelines for Rare Illnesses

Harvard Medical School researchers have developed a new artificial intelligence model, popEVE, that could dramatically accelerate the diagnosis of rare genetic diseases by ranking every variant in a patient’s genome based on its likelihood of causing illness. The tool extends previous work from the Marks Lab at HMS, which created an earlier model called EVE that used evolutionary information to assess how mutations affect protein function. By integrating human population data and a protein large-language model, the new system allows clinicians to compare variants across genes and place them on a continuous spectrum of predicted disease severity — something earlier methods struggled to do.

In a paper published in Nature Genetics, the researchers show that popEVE can reliably distinguish between pathogenic and benign variants, differentiate healthy individuals from patients with severe developmental disorders, and even predict whether a genetic alteration is likely to lead to death in childhood or adulthood. Crucially, the model demonstrated no detectable ancestry bias and avoided overpredicting harmful mutations, a major challenge in existing variant-classification tools. The analysis, according to an article in hms.harvard.edu, also confirmed whether variants were inherited or spontaneous, even in cases where parental data was unavailable.

When applied to a cohort of about 30,000 individuals with severe developmental disorders who lacked prior diagnoses, popEVE identified likely causal variants in roughly one-third of cases. Most strikingly, the model uncovered 123 previously unknown genes linked to developmental disorders, 25 of which have already been independently validated by other laboratories. These discoveries highlight popEVE’s potential to uncover disease-causing mutations that traditional approaches routinely miss, particularly in patients who have long sought answers.

The research team is now making popEVE available to clinicians through an online portal and integrating its scores into major genomic databases such as UniProt and ProtVar. Early clinical collaborations — including with Boston Children’s Hospital, the Children’s Hospital of Philadelphia, and Genomics England — suggest popEVE is already helping physicians interpret complex genomic data and deliver long-delayed diagnoses to patients with rare conditions. While further validation is needed before widespread clinical adoption, researchers believe popEVE could eventually reshape genetic medicine by streamlining diagnoses, guiding treatment decisions, and identifying entirely new drug targets.

read more at hms.harvard.edu