Imagine a world where diagnosing rare diseases becomes faster, simpler, and more accessible, even in under-resourced healthcare settings. That's the promise of a groundbreaking AI model called popEVE, which leverages the power of evolution to revolutionize genetic interpretation. But here's where it gets controversial: can an algorithm truly outsmart the complexities of the human proteome? And this is the part most people miss—it's not just about identifying harmful mutations; it's about doing so equitably, ensuring no patient is left behind due to underrepresentation in genetic databases.
Missense variants have long been a thorn in the side of geneticists, their subtle and context-dependent effects making them notoriously difficult to interpret. While existing prediction models excel in well-studied disease genes, they often stumble in the uncharted territories of the proteome. Enter popEVE, a deep generative model developed by researchers from Harvard Medical School and the Center for Genomic Regulation (CRG) in Barcelona. Published in Nature Genetics under the title ‘Proteome-wide model for human disease genetics’, this tool combines evolutionary data with human population datasets to assess variant deleteriousness across the entire proteome.
What sets popEVE apart is its ability to operate with minimal input—even a patient’s genetic information alone suffices. This is a game-changer for rare disease diagnosis, where resources are often limited, and time is of the essence. As Mafalda Dias, PhD, co-corresponding author of the study, explains, ‘Clinics don’t always have access to parental DNA, and many patients come alone. popEVE can help these doctors pinpoint disease-causing mutations, and we’re already seeing its impact through collaborations with clinics.’
The sheer scale of disease-causing genetic variation is daunting, far too vast to be tackled by population studies or experimental assays alone. But the biodiversity of life on Earth offers a unique lens, revealing genetic patterns preserved over billions of years of evolution to maintain fitness. By comparing protein sequences across species, computational models like popEVE can identify amino acid positions critical for life. This approach builds on the team’s earlier work, EVE (Evolutionary model of Variant Effect), which classified mutations in human disease genes as benign or harmful but lacked comparability across genes.
popEVE bridges this gap by integrating evolutionary data with insights from the UK Biobank and gnomAD, databases that provide information on variants present in healthy individuals. This calibration ensures the model’s scores are not only accurate but also comparable across genes. To test its mettle, the researchers analyzed genetic data from over 31,000 families with children affected by severe developmental disorders. In 98% of cases where a causal mutation was known, popEVE correctly ranked it as the most damaging variant in the child’s genome, outperforming even DeepMind’s AlphaMissense.
But popEVE didn’t stop there. It identified 123 new candidate disease genes previously unlinked to developmental disorders, many of which are active in the developing brain and interact with known disease proteins. Strikingly, 104 of these genes were observed in just one or two patients, highlighting the model’s ability to uncover rare yet critical genetic insights.
Perhaps most importantly, popEVE addresses a long-standing issue in genetics: underrepresentation in global databases. As Jonathan Frazer, PhD, co-corresponding author, notes, ‘No one should get a scary result just because their community isn’t well represented in global databases. popEVE helps fix that imbalance, something the field has been missing for a long time.’
But here’s the question that lingers: As we rely more on AI to decode the complexities of human genetics, are we risking oversimplification? Could popEVE’s success in rare disease diagnosis overshadow the need for diverse, inclusive genetic data? We’d love to hear your thoughts—do you think popEVE is a leap forward, or does it raise more questions than it answers? Share your perspective in the comments below!
