AI-driven temporal decoding of endothelial mechanotransduction and its medico-engineering applications in assessing pan-cancer prognosis

NIU Niu; WU Bin; LIU Hu; WANG Ning; ZHANG Ke; HUANG Sizhao

doi:10.3969/j.issn.1674-3806.2026.01.04

Abstract：Objective To decode the temporal transcriptional responses of endothelial cells to distinct shear stress patterns,and to construct a low fluid shear stress(LFSS)molecular features,and to assess its clinical value in pan-cancer prognosis,and to explore the application potential of artificial intelligence(AI)for dynamic mechanobiological prediction.Methods Three independent human endothelial transcriptome datasets related to shear stress were integrated by using the Gene Expression Omnibus(GEO).Robust differentially expressed genes(DEGs)were identified using cross-cohort meta-analysis by weighted Stouffer's Z method.Temporal expression patterns under steady shear stress(ST),oscillatory shear stress(OS),and pulsatile shear stress(PS)were decoded by using Theil-Sen regression and k-means clustering.LFSS scoring was constructed based on LFSS-specific upregulated genes and was evaluated in The Cancer Genome Atlas(TCGA)pan-cancer atlas cohort(n=11 160)by using Kaplan-Meier survival analysis and multivariable Cox proportional hazards regression models.A long short-term memory(LSTM)network model was further constructed to predict shear stress-induced gene expression dynamics and compared with random forest(RF)and support vector regression(SVR)models.Results A total of 811 robust DEGs were identified across the datasets.Temporal analysis revealed that OS induced aberrant cell cycle and DNA replication programs via the sustained activation of the YAP/TAZ signaling axis,whereas PS predominantly triggered a physiological protective response via the KLF2/KLF4 pathway.The LFSS scores were significantly correlated with overall survival in13 types of cancers[false discovery rate(FDR)＜0.05]and were positively correlated with hypoxia and pathological angiogenesis pathways in the tumor microenvironment.The LSTM model outperformed RF and SVR models,with a coefficient of determination(R2)of 0.842 and a mean absolute error(MAE)of 0.068,demonstrating that the LSTM model had superior generalization performance.Conclusion LFSS can induce specific endothelial transcriptional remodeling and has significant prognostic implications in a variety of cancers.LSTM-based AI models can effectively capture the shear stress-related temporal dynamics,providing a novel medico-engineering framework for decoding vascular mechanical abnormalities and supporting the assessment of precise treatment for tumors.