Most genetic mutations have unknown effects; an AI began to change that. In 2023 a model classified millions of possible missense variants as likely benign or harmful. It offered a map for interpreting the human genome’s variation.
Missense at scale
The target is vast. The tool assessed 71 million possible single-letter protein changes. Coverage is comprehensive.
Structure-informed
Biology guides it. The model builds on protein-structure prediction to judge impact. Physical plausibility matters.
Clinical relevance
Diagnosis benefits. Many variants of uncertain significance received a prediction. Genetic testing gains context.
Not a verdict
Caution applies. Predictions are probabilistic, not definitive diagnoses. Experimental confirmation still matters.
Open catalog
Access is broad. The predictions were released for researchers. Uptake was immediate.
A growing field
Others followed. Variant-effect prediction is now a busy research area. Methods keep improving.
The bottom line
AlphaMissense predicted the harm of 71 million protein variants using structure-informed learning. It gives context to genetic variants of uncertain significance. It advanced genome interpretation.
