What AlphaMissense predicts
AlphaMissense estimates whether an amino acid substitution is likely to damage protein function. It uses protein structure and evolutionary data instead of human clinical labels, so it can still score rare variants that are absent from curated databases.
How AlphaMissense is different
REVEL and DANN use existing annotations, population data, and other prediction outputs. AlphaMissense starts with the changed amino acid, the protein position, and conservation across species. It does not train directly on human clinical databases or labels.
Where it helps
Database-trained models are strongest when similar variants have already been observed and labeled. AlphaMissense can still score a novel missense variant because it evaluates the protein position and amino acid substitution.
Classification ranges
The AlphaMissense paper groups scores into three ranges. The likely benign and likely pathogenic cutoffs were chosen to reach 90% precision in ClinVar benchmarks.
- Less than 0.34: likely benign.
- 0.34 to 0.564: ambiguous.
- Greater than 0.564: likely pathogenic.
In Gene Inspector Pro
Gene Inspector Pro shows the value as AMS next to other predictor values. The High AlphaMissense Score panel includes variants with score >= 0.5 so they can be reviewed in one place.
How to use it
Treat AlphaMissense as one review signal, not a diagnosis. Check it against variant consequence, allele frequency, ClinVar records, functional hotspots, genotype quality, inheritance, and the question being reviewed.