Unraveling miRNA Networks in Biological Aging: Machine Learning Analysis of 1300 human miRNA Profiles to Enhance Vascular Regeneration Potential

Aging is the strongest independent risk factor for cardiovascular disease, driven by complex, multifactorial processes involving senescence, fibrosis, apoptosis, and vascular remodeling. MicroRNAs (miRNAs) have emerged as central regulators of these age-associated epigenetic changes. Although individual miRNAs are implicated in cellular aging features, comprehensive insight into miRNA-driven biological aging remains limited. Here, we employ machine learning on 1300 miRNA sequencing samples from patient tissue to explore the intricate miRNA alterations linked to aging and identify potential therapeutic targets.

Our dataset includes over 1300 non-tumor control tissue samples from The Cancer Genome Atlas (TCGA), covering patients aged 20 to 90 years across 13 tissue types. Patients were categorized into age groups based on clinical geriatric definitions. Initial data exploration included t-SNE, ANOVA, and subgroup analysis, followed by machine learning classification (SMOTE-balanced, 80/20 train-test split, and 5-fold cross-validation) using decision tree models and random forest classifiers. Accuracy was evaluated via confusion matrices and standard performance metrics, while miRNA impacts were assessed through variable importance plotting (VIP). Network analyses were performed using miRNetR and gene ontology enrichment for pathway analysis. R and Python were used for all data processing.

Our dataset’s age distribution was non-normal (p<0.5, Shapiro-Wilk) with a mean age of 62. In t-SNE analysis, miRNA profiles showed distinct tissue specificity but less pronounced age differentiation. Across samples, 183 miRNAs were significantly downregulated (p<0.05) and 28 upregulated (p<0.05) in older vs. younger patients. Tissue-specific miRNA regulation highlighted distinct networks (p<0.01), with few miRNAs consistently expressed age-dependently across up to 6tissues. In the four most common tissues (breast, kidney, liver, lung), five miRNAs showed consistent downregulation in aging (p<0.05). Our models achieved age-group classification accuracies ranging from 0.52 (dummy classifier) to 0.85 (random forest) with a precision of 0.83 and recall of 0.85. Variable importance plotting of the random forest classification algorithm showed a Mean Decrease accuracy of single miRNAs of up to 6.5%. All miRNAs with a Mean Decrease accuracy of over 3.25% were considered highly systemically relevant, leaving 14 significantly upregulated miRNAs and 37 significantly downregulated miRNAs with high systemical relevance. These miRNAs were used to create regulatory networks of target genes which are significantly overregulated in aging (p<0,05) and aging-associated cellular processes, like proliferation, drug response and cell differentiation (p<0,05). Centrally involved miRNAs in these networks, particularly miR-375, miR-34a, and miR-148a, showed significant upregulation in replicative senescent endothelial cells (qPCR, p<0.05).

Our findings suggest that miRNA regulation in aging extends beyond simple causality, with tissue- and cell-specific factors shaping these complex networks. Given the strong contribution of endothelial cells, due to their number, to tissue miRNA profiles, we hypothesize these networks are relevant to vascular aging and senescence. We aim to leverage these insights to develop miRNA-based agents for reprogramming aging-associated gene networks, potentially enhancing endothelial function and vascular regenerative capacity after injury.