Pathogenicity Prediction of GABA_AReceptor Missense Variants

Abstract

Variants in the genes encoding the subunits of gamma-aminobutyric acid type A (GABAA) receptors are associated with epilepsy. To date, over 1000 clinical variants have been identified in these genes. However, the majority of these variants lack functional studies and their clinical significance is uncertain although accumulating evidence indicates that proteostasis deficiency is the major disease-causing mechanism for GABAAreceptor variants. Here, we apply two state-of-the-art modeling tools, namely AlphaMissense, which uses an artificial intelligence-based approach based on AlphaFold structures, and Rhapsody, which integrates sequence evolution and known structure-based data, to predict the pathogenicity of saturating missense variants in genes that encode the major subunits of GABAAreceptors in the central nervous system, includingGABRA1,GABRB2,GABRB3, andGABRG2. Our results demonstrate that the predicted pathogenicity correlates well between AlphaMissense and Rhapsody although AlphaMissense tends to generate higher pathogenic probability. Furthermore, almost all annotated pathogenic variants in the ClinVar clinical database are successfully identified from the prediction, whereas uncertain variants from ClinVar partially due to the lack of experimental data are differentiated into different pathogenicity groups. The pathogenicity prediction of GABAAreceptor missense variants provides a resource to the community as well as guidance for future experimental and clinical investigations.