Researchers from Harvard Medical School and the Center for Genomic Regulation in Barcelona have developed an AI model, popEVE, that can identify and rank the severity of disease-causing genetic mutations, even those never before seen in humans. Published in the journal Nature Genetics, this tool analyzes evolutionary data from hundreds of thousands of species combined with human population genetics to determine which parts of our roughly 20,000 proteins are essential for life.
This allows popEVE to perform a critical task: directly compare the predicted severity of a mutation in one gene against a mutation in another gene across the entire human genome. In validation tests using data from over 31,000 families with severe developmental disorders, popEVE correctly identified the known causal variant as the most damaging in the child's genome in 98% of cases. The researchers note it outperformed other models, including DeepMind's AlphaMissense.
A significant advantage of popEVE is its potential to reduce racial bias in genetic diagnosis. Unlike tools that rely heavily on existing human databases—which are skewed toward European ancestry—popEVE's evolution-based core treats all human variants equally, leading to fewer false positives for people from underrepresented groups. The tool is particularly poised to accelerate diagnoses for rare diseases, as it can work with a single patient's genetic data, making it a powerful and cost-effective solution for clinics with limited resources.