A novel multi-omics–machine learning pipeline reveals immune and metabolic links between type 2 diabetes and atherosclerosis

Background Atherosclerosis (AS) and type 2 diabetes mellitus (T2DM) frequently coexist, jointly accelerating cardiovascular complications through shared inflammatory and metabolic pathways. Despite extensive research, the molecular mechanisms linking these chronic diseases remain incompletely defined. Purpose This study aimed to delineate the shared transcriptional signatures and identify candidate biomarkers contributing to T2DM-associated AS progression using an integrative multi-omics strategy. Research Design A retrospective bioinformatics investigation integrating differential expression analysis, co-expression network modeling, protein-interaction profiling, immune deconvolution, and machine-learning–based biomarker prioritization was conducted.Study Sample: Publicly available transcriptomic datasets were obtained from the NCBI Gene Expression Omnibus, including AS tissue samples (GSE100927), pancreatic islet samples from individuals with T2DM (GSE25724), and two independent datasets for external validation (GSE30169 and GSE26168). Data Collection and/or Analysis Differentially expressed genes (DEGs) were identified for AS (n = 3,368) and T2DM (n = 4,553). DEG intersection and Weighted Gene Co-expression Network Analysis (WGCNA) revealed 443 shared crosstalk genes. Enrichment analyses highlighted immune activation processes (e.g., leukocyte-mediated immunity, lysosomal pathways) and metabolic dysregulation (e.g., mitochondrial-mediated apoptosis, TGF-β signaling). A protein-protein interaction network was constructed, identifying high-degree hub genes such as HLA-DRB1, JAK3, and MFN1. Immune cell profiling using CIBERSORTx compared disease microenvironments, demonstrating increased M1 macrophages and helper T cells in AS, and elevated monocytes and B cells in T2DM (p < 0.05). A fine-tuned TabNet model ranked predictive biomarkers (e.g., BTK, ZAP70, CD4) showing strong diagnostic performance (AUC = 0.85 for AS; 0.79 for T2DM). Results The integrative multi-omics workflow uncovered a robust set of immune-metabolic crosstalk genes shared between AS and T2DM. Hub-gene analysis and immune infiltration patterns demonstrated convergent dysregulation in lysosomal activity, mitochondrial integrity, and adaptive immune signaling. Machine-learning prioritization further identified a subset of biomarkers capable of discriminating disease states with high accuracy, strengthening their translational potential. Conclusions This study provides a comprehensive molecular framework linking T2DM and AS, revealing previously unrecognized lysosomal and mitochondrial pathways that may drive their synergistic pathology. The identified biomarkers and immune signatures offer promising avenues for early diagnosis and targeted therapeutic development in patients with comorbid T2DM and atherosclerosis.