Career Center AES Connect

Abstracts

GABAA Receptor Functional Variant Effect Prediction Using Multi-Task Phenotypic Learning (GENTLY)

Abstract number : 2.052
Submission category : 12. Genetics / 12A. Human Studies
Year : 2025
Submission ID : 702
Source : www.aesnet.org
Presentation date : 12/7/2025 12:00:00 AM
Published date :

Authors :
Presenting Author: Christian Bosselmann, MD – Department of Neurology and Epileptology, Hertie Institute for Clinical Brain Research, University of Tuebingen, Tuebingen, Germany

Rebekka Staal Dahl, MSc – Department of Epilepsy Genetics and Personalized Medicine, Danish Epilepsy Center, Dianalund, Denmark
Sebastian Ortiz, MD – Department of Epilepsy Genetics and Personalized Medicine, Danish Epilepsy Center, Dianalund, Denmark
Dennis Lal, PhD – Department of Neurology, McGovern Medical School, The University of Texas Health Science Center at Houston, Houston, TX, USA
Holger Lerche, MD – Department of Neurology and Epileptology, Hertie Institute for Clinical Brain Research, University of Tuebingen, Tuebingen, Germany
Nico Pfeifer, PhD – Department of Computer Science, Methods in Medical Informatics, University of Tuebingen, Tuebingen, Germany
Jules Kreuer, MSc – Department of Computer Science, Methods in Medical Informatics, University of Tuebingen, Tuebingen, Germany
Nathan Absalom, PhD – School of Medical Sciences, Faculty of Medicine and Health, Brain and Mind Centre, The University of Sydney, Sydney, NSW, Australia
Philip Ahring, PhD – School of Medical Sciences, Faculty of Medicine and Health, Brain and Mind Centre, The University of Sydney, Sydney, NSW, Australia
Rikke Moller, PhD – Danish Epilepsy Center Filadelfia

Rationale: γ‐Aminobutyric acid type A (GABAA) receptors are critical for inhibitory neurotransmission. Variants in these receptors are involved in the pathophysiology of both common and rare epilepsy syndromes. Variant effects on channel biophysical function, classified as gain-of-function (GOF) or loss-of-function (LOF), are associated with key clinical characteristics including treatment response. Understanding and predicting variant effects is therefore essential to improve the care for individuals with GABAA-related disorders.

Methods: We collected clinical data from 505 individuals with 272 GABAA receptor variants (174 unique variants). All variants were evaluated with in vitro electrophysiology. Variants were annotated with features based on sequence (e.g., physicochemical properties, conservation), structure (e.g. binding sites, domains), and phenotype. We trained multi-task multi-kernel learning (MTMKL) SVMs on all features (full model) and without clinical features. Model performance was estimated using repeated k-fold cross-validation and ablation.

Results: Our model enables highly accurate prediction of variant functional effects in GABAA receptors (full model: AU-ROC 0.938±0.036; model without clinical features: AU-ROC 0.842±0.060; two-sample t-test: p < 0.0001), outperforming state-of-the-art genome-wide predictors (LoGoFunc: AU-ROC 0.495; evo2-40B: AU-ROC 0.559). Model scores correlated strongly with EC50 (GABA sensitivity; Pearson correlation coefficient: R = -0.76, p < 0.001). Predictions were consistent with expert-based structure-function hypothesis: variants located in domains M1 and M2 were more likely to be predicted as GOF (p < 0.0001), and variants in GABA binding sites were more likely to be predicted as LOF (p = 0.049). Predictions on variants from population databases behaved as expected: 13,389 population variants were similar to functionally neutral variants, and predictions from 87 (likely) pathogenic ClinVar variants were similar to LOF/GOF variants. Under the assumption that GENTLY can provide functional evidence (ACMG PS3 criterion), 9.5-48.4% of variants of unknown significance in ClinVar could be reclassified depending on the choice of score threshold. Lastly, we show that a simple k-nearest neighbor algorithm can predict likely clinical characteristics (median Lin similarity 0.754 IQR 0.161; random baseline median 0.506 IQR 0.160; two-sample t-test: p < 0.0001).
Genetics