AI SCIENCE
This study might change how doctors read your genome
Researchers have developed a new AI model, popEVE, that can identify which mutations in human proteins are most likely to cause disease, even if they’ve never been seen before.
It uses genetic data from hundreds of thousands of species and human populations to understand which parts of proteins are essential for life and which can tolerate change.
This makes popEVE useful for diagnosing rare diseases, where half of the patients never receive a clear explanation.
Because it works with the patient’s genome alone, it helps doctors prioritise the most damaging variants quickly, even when family DNA isn’t available.
Existing tools can predict whether a mutation is harmful, but many cannot compare severity across different genes.
popEVE solves this by combining evolutionary data with human datasets like the UK Biobank and gnomAD. It can now rank mutations across all ~20,000 human proteins on the same scale.
In tests with over 31,000 families affected by severe developmental disorders, popEVE correctly identified the known disease-causing mutation as the most damaging in 98% of cases.
Here’s what you should know:
popEVE ranks mutations across all human proteins on one comparable scale.
It performs strongly for rare, one-off mutations with no case history.
It reduces bias for underrepresented populations in genetic databases.
It also uncovered 123 potential new disease genes, many active in early brain development.
A key strength is that popEVE reduces bias: it treats all human variants equally, avoiding false positives that occur when tools rely too heavily on European genetic datasets.
The researchers note that popEVE only assesses mutations that change proteins and should be used alongside clinical judgement, not as a replacement.
Permission to sequence me. - MV