AlphaMissense

AlphaMissense is an AI model developed by Google DeepMind that predicts the pathogenicity of all possible 71 million human missense variants (where a single letter in the DNA code changes a single amino acid in a protein).

Key Capabilities

• Comprehensive Coverage: It has classified 89% of all possible human missense variants as either likely benign or likely pathogenic, vastly expanding upon the 0.1% of variants whose effects have been experimentally confirmed by human experts.
• Structural Intuition: AlphaMissense builds upon the AlphaFold architecture. It is uniquely able to “understand” how a specific single-amino-acid change might disrupt the 3D folded structure or active binding site of the resulting protein.
• Clinical Utility Base: Its pre-computed predictions are available in standard Variant Call Format (VCF), making it trivial to integrate into existing clinical variant annotation pipelines (like Ensembl VEP) to assist in diagnosing rare genetic diseases.

Usage in Healthcare

When a patient with an undiagnosed rare disease has their genome sequenced, doctors often find thousands of “Variants of Uncertain Significance” (VUS) – mutations never seen before. AlphaMissense provides an immediate, highly accurate algorithmic prediction of whether each mutation is harmless or disease-causing, helping geneticists prioritize which mutations to investigate further.