Precision diagnosis of GABRA1-associated encephalopathies and epilepsy: optimizing variants classification and molecular subregional effects

BackgroundGABRA1 variants are associated with a broad spectrum of epileptic phenotypes ranging from mild idiopathic generalized epilepsy to severe developmental and epileptic encephalopathy (DEE). To date, the majority of the identified GABRA1 variants are missense. Evaluating the pathogenicity of missense variants is a great challenge in genetics. This study aimed to explore reliable biological tools to optimize pathogenic classification of variants, thereby improving precision diagnosis of GABRA1-associated encephalopathies and epilepsy.MethodsThe dataset of disease-associated and control GABRA1 missense variants was curated. The location of these variants was visualized, to analyze the molecular subregional effects. The performance of 34 algorithms in evaluating the pathogenicity of GABRA1 variants was systematically analyzed, including accuracy, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), Matthews correlation coefficient (MCC), F-score, and the area under the receiver operating characteristic curve (AUC).ResultA total of 61 GABRA1 missense variants were analyzed, including 30 pathogenic/likely pathogenic variants from patients with GABRA1-associated epilepsies and 31 benign/likely benign controls from the gnomAD database. The pathogenicity and phenotypes of these variants showed significant domain dependence: all transmembrane variants caused severe developmental and epileptic encephalopathy (DEE), the extracellular domain had the highest phenotypic heterogeneity, and the phenotype distribution differed significantly between functionally critical regions and other regions (P = 0.01), indicating a molecular subregional effect. We evaluated 34 commonly used algorithms, which varied considerably in performance. Ensemble and deep learning algorithms showed superior overall performance, with MetaLR and PrimateAI achieving the highest accuracy (0.9167) and AlphaMissense yielding the best AUC (0.9644). Tools like M-CAP and CADD_phred had low specificity. All tools except fathmm-XF showed highly significant score differences between groups (P < 0.0001), and high-performance tools presented a clear bimodal distribution with minimal overlap.ConclusionEnsemble learning and deep learning algorithms are highly effective for predicting the pathogenicity of GABRA1 missense variants. These computational tools provide reliable support for the pathogenicity assessment of GABRA1 variants in clinical genetic diagnosis.