Multi-Omics Integration with Machine Learning for Early Detection of Ischemic Stroke Through Biomarkers Discovery

Abstract

A complicated and diverse neurological condition, ischemic stroke (IS) is marked by a high death rate and substantial long-term impairment.  Despite extensive research, reliable biomarkers for the clinical diagnosis and prognosis of ischemic stroke remain elusive, and the underlying molecular pathways remain obscure.  This study uses a physiologically informed Convolutional Neural Network (BioCNN) to integrate multi-omics and present a novel method for the early identification and classification of ischemic stroke subtypes. Multi-omics Thirty acute ischemic stroke patients who were hospitalized within twenty-four hours after the beginning of symptoms provided data. Multi-omics profiling included mRNA, miRNA, circRNA, and DNA methylation datasets. After rigorous preprocessing, the integrated into a biomedical knowledge graph to enable graph-based learning. The BioCNN model outperformed models based on individual omics layers in terms of prediction performance.  Its accuracy was 97.89%, its F1-score was 96.48%, and its AUC was 95.12%. Comparative analyses also revealed that among single-omics models, mRNA data yielded the best results, highlighting the complementary value of multi-omics integration. These findings emphasize the effectiveness of deep learning frameworks combined with integrated multi-omics data for advancing biomarker discovery and accurate classification of ischemic stroke subtypes, offering promising implications for early diagnosis and personalized treatment strategies.