Back

PrimateAI-3D outperforms AlphaMissense in real-world cohorts

Parry, D. A.; Bosc, T.; Hamp, T.; Fiziev, P.; Sharma, A.; Kassam, I.; McRae, J.; Farh, K.

2024-01-13 genetic and genomic medicine
10.1101/2024.01.12.24301193 medRxiv
Show abstract

Accurately predicting the impact of genetic variants is essential for interpreting genomic data, yet no consensus exists on how to measure classifier performance. We prepared the most comprehensive set of benchmarks to date and applied them to the recently published models PrimateAI-3D and AlphaMissense. PrimateAI-3D outperforms AlphaMissense on rare-disease cohort and biobank benchmarks, indicating that performance on clinical databases or in vitro assays does not reliably generalize to real-world cohorts.

Matching journals

The top 2 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.