Choudhary, D.; Portelli, S.; Ascher, D. B.

Variants of uncertain significance (VUS) in USH2A represent a critical interpretive challenge in inherited retinal disease, with over 70% of ClinVar submissions for this gene currently unresolved. We developed UshEffect-3D, a gene-specific, structure-informed machine learning framework for USH2A missense variant classification. A dataset of 545 curated variants was assembled from ClinVar and LOVD, and AlphaFold2-predicted domain structures were used to generate local structural descriptors combined with sequence-based evolutionary conservation scores, yielding nine features after sequential selection. Eleven classifiers were trained using 10-fold cross-validation and evaluated on a blind test set and 78 ACMG-classified pathogenic variants. The Random Forest classifier achieved an MCC of 0.87 and AUC of 0.97 on the blind test set, substantially outperforming general-purpose predictors including PolyPhen-2 (MCC = 0.61), AlphaMissense (MCC = 0.42), and ESM-1b (MCC = 0.32). SHAP and ablation analysis identified evolutionary conservation as the dominant predictor, with structural stability providing an independent complementary signal. Applied to 2,639 ClinVar VUS, the model prioritised 33.6% as likely pathogenic, with enrichment in the Laminin N-terminal and Laminin G-like domains. UshEffect-3D provides a high-confidence prioritisation resource for the large unresolved VUS burden in USH2A and is freely accessible via an interactive web server.