Researchers built Varformer, an open-source AI tool that ranks likely drug targets by combining DNA spelling changes with gene-level features. Trained on clinically validated targets across four ancestry groups, it scored up to 0.95 precision and beat rival methods by 20–23% on the hardest benchmark.
Finding drug targets is slow because validated targets are scarce and models struggle to combine variants from many people. Varformer learns across four ancestry groups, so it can flag targets with genetic support across populations. It also scores single variants — ranking a carrier's known disease variant in their top 20 results 57% of the time.
✦ Better tools for reading our genes could bring new medicines to more people, sooner.