DITTO: An explainable machine-learning model for transcript- specific variant pathogenicity prediction

Abstract Accurate classification of genetic variants is critical for medical decision-making, providing insights into disease mechanisms, and enabling therapeutic discovery. While numerous methods exist, they often address limited variant types, lack transcript awareness, and operate as opaque black boxes. To address these issues, we developed DITTO - a unified, explainable, transcript-aware advanced machine-learning method for pathogenicity prediction. It integrates diverse genomic features-including conservation scores, population frequencies, etc., to train a single, explainable neural network model. DITTO outperforms existing methods across standard benchmarks, demonstrating superior performance (99% F1 score) in classifying pathogenic and benign variants. DITTO is publicly available at https://github.com/uab-cgds-worthey/DITTO