Functional Effect Predictions For Ion Channel Missense Variants Using a Protein Language Model

Abstract

Channelopathies represent a group of diseases often caused by missense variants in ion channels affecting the functioning of tissues like the nervous system, heart, and muscle. The gold standard for functionally characterizing a variant is to measure the electrophysiological changes in channel properties using cell-based heterologous expression systems. As this method is time-consuming and generally unavailable, clinical practice often relies on in silico models to predict the functional consequences of ion channel variants. We constructed a Missense ION (MissION) channel variant classifier based on a protein language model and trained on 3176 gain- or loss-of-function variants, the largest set collected to date, in order to predict the functional effects of variants across a broad range of ion channels. MissION achieves a significant increase in predictive performance (Area Under the Receiver Operating Characteristic Curve (ROC-AUC): 0.925, compared to 0.897 for the current leading model). Moreover, the model generalizes well to ion channel genes for which little or no electrophysiological recordings are available. MissION provides functional predictions for over 600,000 ion channel variants, made available through an online interface at www.synaptica.nl/variant-interpreter that allows variant interpretation for a wide range of channelopathies.