AI ‘popEVE’ Learns from Evolutionary History to Accelerate Rare Disease Diagnosis
Table of Contents
A groundbreaking artificial intelligence model, popEVE, is poised to revolutionize teh diagnosis of rare genetic diseases by leveraging the vast record of evolution and offering a more equitable approach to identifying harmful mutations.
Nearly half of individuals living with a rare disease currently lack a definitive diagnosis, a statistic researchers hope to dramatically improve with this new tool. Developed by scientists at Harvard Medical School and the Center for Genomic Regulation (CRG) in Barcelona, popEVE doesn’t just identify disease-causing mutations – it ranks their severity, even for mutations never before observed in any person. The findings, published today in Nature genetics, represent a significant leap forward in precision medicine.
The Power of Evolutionary Insight
Traditional methods of identifying harmful genetic mutations often rely on comparing patterns across large groups of patients. However, this approach falters when dealing with truly rare conditions where sufficient data is lacking. popEVE takes a different tack, drawing on the principle that mutations that have persisted throughout evolution are less likely to be damaging. The original EVE model, developed by the same team, demonstrated that AI could predict the effects of mutations as or better than many lab-based experiments. However, EVE’s scores weren’t directly comparable across different genes, hindering its ability to identify the most damaging mutation within a patient’s genome.
popEVE: Ranking Mutations Across the Human Proteome
popEVE builds upon the foundation of EVE by incorporating data from the UK Biobank and gnomAD, two extensive repositories of human genetic facts. This allows the model to calibrate its predictions based on the prevalence of variants in healthy individuals.
The result is the first AI capable of ranking mutations across the entire human proteome – the complete set of roughly 20,000 proteins encoded within the human genome. “A mutation in gene A can now be compared directly with one in gene B on the same severity scale,” explained a senior official involved in the research.This capability is crucial for clinicians, enabling them to prioritize the most perhaps damaging variants for further examination.
To validate popEVE’s performance, researchers analyzed genetic data from over 31,000 families affected by severe developmental disorders. In an impressive 98% of cases where a causal mutation was already known, popEVE correctly identified it as the most damaging variant. The model also outperformed leading competitors, including DeepMind’s AlphaMissense. Moreover, popEVE uncovered 123 previously unknown candidate disease genes, many of which are active in the developing brain and interact with known disease proteins.
Addressing Bias and Improving Accessibility
A key strength of popEVE lies in its ability to mitigate bias inherent in many existing genetic tools. These tools often flag variants as potentially harmful simply because they haven’t been observed before, disproportionately affecting individuals from underrepresented ancestral groups.
“no one should get a scary result just because their community isn’t well represented in global databases,” stated a researcher at the Center for Genomic Regulation. “popEVE helps fix that imbalance,something the field has been missing for a long time.” By treating all human variants equally, regardless of their frequency in specific populations, popEVE significantly reduces false positives.
The model’s accessibility is another significant advantage. It requires only a patient’s genetic information, eliminating the need for parental DNA samples, which are frequently enough arduous to obtain.This is particularly beneficial in healthcare systems with limited resources, where faster, simpler, and cheaper diagnoses are critical. “Clinics don’t always have access to parental DNA and many patients come alone. popEVE can help these doctors identify disease-causing mutations, and we’re already seeing this from collaborations with clinics,” noted a co-corresponding author of the study.
While popEVE represents a major advancement, researchers emphasize that it is not a replacement for clinical judgment. The model interprets only DNA changes that alter proteins, and doctors must still consider medical histories and symptom analysis to arrive at an accurate diagnosis.
Further information about the study can be found in the Nature Genetics article, “Proteome-wide model for human disease genetics” (DOI: 10.1038/s41588-025-02400-1). This innovative AI promises to bring hope to countless individuals and families affected by the challenges of rare disease diagnosis.
