Standing
Provided proteome-wide predictions for human missense variants, a scale that experimental characterisation cannot reach.
Variant effect · DeepMind · 2023
AlphaMissense
Classifies missense variants as likely benign or likely pathogenic across the proteome, adapting an AlphaFold-derived model with population frequency data rather than clinical labels.
In plain language
Given one amino-acid substitution, it estimates how likely that change is to break the protein. It was trained without clinical labels, which is what makes testing it against clinical data meaningful.
One way to picture it
A typo checker that knows which letters matter. Changing one in a load-bearing word is not the same as changing one in a filler word.
Commonly misread as
A score is one line of computational evidence inside a variant interpretation framework. It is not a diagnosis and it does not cover insertions or deletions.
Distinctions
- Trained without clinical annotations, which keeps the evaluation against clinical databases meaningful.
- Uses structural context, so it separates buried positions from exposed ones.
- Outputs a calibrated score rather than a binary call.
Limits
A score is computational evidence within a variant interpretation framework, not a diagnosis. Its predictions concern single amino-acid substitutions only.
Weights and code
Checked against the registry, not from memory| Repository | Size | Licence | Note |
|---|---|---|---|
| google-deepmind/alphamissense | Predictions released as tables | CC BY-NC-SA 4.0 | — |