Evo 2 BRCA1 Variants
Scoring BRCA1 variants for likely harm with the Evo 2 genomic model, with no training, and reporting where it fails as carefully as where it works.
Spec
- Model
- Evo 2 genomic foundation model
- Method
- Zero-shot Δ log-likelihood
- Benchmark
- Findlay 2018, ~3,893 SNVs
- Also checked
- ClinVar
- AUROC
- 0.737 (loss-of-function vs functional)
- AUPRC
- 0.564 (baseline 0.23)
- Compute
- Modal
Write-up coming soon.
Why I built it
Who it was for and what problem they had.
The hard part
One decision I had to make and what I gave up.
What I'd do next
Honest limits and the next measurement.