Integrating Transcriptomics and Epigenomics with Co-Attention for Mild Cognitive Impairment Progression Prediction

Abstract:
Determining whether mild cognitive impairment (MCI) will progress to Alzheimer's disease (AD) is crucial for the early diagnosis and intervention of AD. This study proposes a novel co-attention-based model that integrates transcriptomic and epigenomic data to predict MCI progression. By using gene expression to guide the selection of relevant DNA methylation features, the model captures both gene activity and potential regulatory mechanisms, which are crucial for understanding how gene expression is regulated in diseases. Unlike traditional multi-omics integration methods that rely on direct feature concatenation, our approach prioritizes transcriptomics, aligning with its downstream role in gene regulation and enhancing interpretability. The model was validated on two independent cohorts from the Alzheimer’s Disease Neuroimaging Initiative (ADNI), demonstrating superior predictive performance compared to conventional models (AUROC = 0.931, accuracy = 0.853, sensitivity = 0.847 and specificity = 0.838). These findings suggest that integrating transcriptomics and epigenomics using a co-attention mechanism provides a more comprehensive framework for understanding MCI progression, potentially uncovering key molecular pathways and biomarkers for early intervention and therapeutic development.