Functional effect predictions for ion channel missense variants using a protein language model.

Area of Science:

• Genetics
• Computational Biology
• Biophysics

Background:

• Channelopathies are diseases caused by ion channel dysfunction, often due to missense variants.
• Electrophysiological studies are the gold standard for variant characterization but are time-consuming and inaccessible.
• In-silico models are increasingly used to predict variant effects in clinical practice.

Purpose of the Study:

• To develop a novel computational model for predicting the functional consequences of ion channel missense variants.
• To create the largest dataset of ion channel variants for training and validation.
• To provide a widely accessible tool for interpreting variant effects in channelopathies.

Main Methods:

• Construction of a protein language model-based classifier named MissION.
• Training the model on a dataset of 1996 gain- or loss-of-function variants.
• Evaluating predictive performance using Area Under the Receiver Operating Characteristic Curve (ROC-AUC).

Main Results:

• MissION achieved a high predictive performance (ROC-AUC: 0.918), outperforming existing leading models (0.884 and 0.779).
• The model demonstrates strong generalization capabilities, even for genes with limited electrophysiological data.
• Functional predictions for over 600,000 ion channel variants are now available via an online interface.

Conclusions:

• MissION offers a significant advancement in predicting the functional impact of ion channel variants.
• The tool facilitates variant interpretation for a broad spectrum of channelopathies.
• This computational approach enhances the understanding and clinical management of ion channel diseases.