Machine Learning for Gene-Expression-Based Prediction of Alcohol Use Disorder with an Integrate Web-Based Dashboard

Abstract Alcohol use disorder (AUD) is a complex condition associated with molecular alterations in brain regions involved in reward, motivation, and behavioral regulation. Gene-expression profiling can provide valuable molecular information for studying AUD; however, the high dimensionality and correlation structure of transcriptomic data present challenges for predictive modeling and interpretation. This study develops a machine-learning approach for predicting AUD status from gene-expression profiles and introduces GENO, a web-based dashboard designed to support laboratory users in interpreting the resulting predictions. Publicly available bulk RNA-sequencing data from postmortem brain tissue of 114 individuals, including 58 AUD cases and 56 non-AUD controls, were obtained from the Gene Expression Omnibus repository (GSE253155). Gene-expression data were standardized, and dimensionality was reduced using ANOVA F-test and mutual-information-based feature selection. An Elastic Net logistic regression model was then developed to manage high-dimensional and correlated predictors while maintaining model interpretability. On the held-out test set, the model achieved an accuracy of 86%, an F1-score of 0.865, and a receiver operating characteristic area under the curve (ROC-AUC) of 0.895. For each input profile, the model generates a probability-based prediction of membership in the AUD class and estimates gene-level contributions to the prediction. GENO integrates this predictive pipeline into a web-based interface that supports data upload, automated analysis, result visualization, and structured report generation. A preliminary task-based usability assessment with laboratory users further demonstrated the feasibility of completing the platform’s primary workflow. These findings provide preliminary evidence that interpretable regularized machine learning can predict AUD status from postmortem gene-expression profiles within the analyzed dataset.