AI in Variant Interpretation

Identifying a variant is only the first step. Knowing what it does is the real challenge. AI is moving us beyond simple conservation scores to structural and evolutionary modeling.

AlphaMissense

Building on AlphaFold, AlphaMissense predicts the pathogenicity of missense variants by understanding how amino acid changes affect protein structure stability. It has classified 89% of all possible missense variants.

PrimateAI-3D

Illumina's PrimateAI-3D uses natural selection in non-human primates as a training set. If a variant is common in chimps or bonobos, it is likely benign in humans, even if it looks damaging to other tools.