Google's New AI AlphaMissense Classifies Millions of Protein Mutations, Identifying Potential Disease-Causing Variants

Google DeepMind researchers developed an AI called AlphaMissense that identified and classified 71 million protein mutations, finding 1/3 may cause serious diseases.

AlphaMissense builds on DeepMind's AlphaFold, using predicted protein structures to evaluate if mutations will impact function.

The AI was trained on human/primate protein sequences to learn normal vs abnormal mutations. It classified 89% of variants, with 57% benign and 33% likely pathogenic.

Applications could help identify genetic disease genes and mechanisms, particularly for rare diseases, though drugs not directly produced.

Limitations are it evaluates mutations individually, not combinations, and gives probabilities vs definitive classifications. More validation needed.