Leakage-Controlled Benchmarking of Variant Effect Predictors and Development of Gradient-Boosting Consensus Model

A leakage-controlled Benchmark dataset for missense variants shows benchmarked protein variant effect predictors cluster tightly in AUC (0.631–0.758) and drop from clinical to functional annotations.

• Why it matters: Fairly comparing variant effect predictors is hard due to data circularity, human-only clinical benchmarks, and limited access to code and training data that can inflate reported performance.
• What they did: The authors reviewed 18 recent and 3 older VEPs, built a leakage-controlled Benchmark integrating ClinVar and Humsavar clinical variants plus cross-species functional annotations, then evaluated AUC across clinical versus functional splits.
• The result: They trained PredictSNPprot, a gradient-boosting consensus model, reaching AUC 0.82 on a held-out test set, improving over the original PredictSNP meta-predictor by 13.6 and the best single tool by 5.7 percentage points.

AI summary of the abstract · may contain errors · not medical advice · read the paper.