Endothelial cell senescence: A machine learning-based meta-analysis of transcriptomic studies

Numerous systemic vascular dysfunction that leads to age-related diseases is highly associated with endothelial cell (EC) senescence; thus, identifying consensus features of EC senescence is crucial in understanding the mechanisms and identifying potential therapeutic targets. Here, by utilizing a total of 8 screened studies from different origins of ECs, we have successfully obtained common features in both gene and pathway level via sophisticated machine learning algorithms. A total of 400 differentially expressed genes (DEGs) were newly discovered with meta-analysis when compared to the usage of individual studies. The generated parsimonious model established 36 genes and 57 pathways features with non-zero coefficient, suggesting remarkable association of phosphoglycerate dehydrogenase and serine biosynthesis pathway with endothelial cellular senescence. For the cross-validation process to measure model performance of 36 deduced features, leave-one-study-out cross-validation (LOSOCV) was employed, resulting in an overall area under the receiver operating characteristic (AUROC) of 0.983 (95 % CI, 0.952, 1.000) showing excellent discriminative performance. Moreover, pathway-level analysis was performed by Pathifier algorithm, obtaining a total of 698 pathway deregulation scores from the 10,416 merged genes. In this process, high dimensional data was eventually narrowed down to 57 core pathways with AUROC value of 0.982 (95 % CI, 0.945, 1.000). The robust model with high performance underscores the merit of utilizing sophisticated meta-analysis in finding consensus features of endothelial cell senescence, which may lead to the development of therapeutic targets and advanced understanding of vascular dysfunction pathogenesis with further elucidation.