Back

DAVE: how to use explainable AI to interpret missense variants for genome diagnostics based on functional protein modeling

Niemeijer, T.; Mulder, R.; Westers, H.; Jongbloed, J. D. H.; Charbon, B.; Sikkema-Raddatz, B.; Johansson, L. F.; van Gijn, M. E.; van Diemen, C. C.; Hendriksen, D.; Abbot, K. M.; Maassen, W. T. K.; Swertz, M. A.; van der Velde, K. J.

2025-11-27 genetic and genomic medicine
10.1101/2025.11.25.25340947 medRxiv
Show abstract

Diagnostic yield in NGS genome diagnostics is constraint by the high fraction of variants of uncertain significance (VUS), in large part due to insufficient interpretability of missense variation. Existing pathogenicity predictors offer strong performance, but often produce an unexplainable score lacking mechanistic insight. Here, we present the Digital Approxima-tion of Variant Effects (MOLGENIS DAVE), an explainable missense variant predictor built on 12 biophysically grounded features spanning stability, hydrophobicity, electrostatics, and molecular interactions. Trained on curated Dutch diagnostic data, DAVE reliably classifies and breaks down predictions into interpretable feature contributions. With a focus on ex-plainability, this framework aims to alleviate the VUS burden, advances clinically actionable variant interpretation and enables mechanistic follow-up. All source code used to process and integrate the data, along with the data required to reproduce all annotations and anal-yses, is available at https://github.com/molgenis/dave.

Matching journals

The top 8 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.