Population cohort-validated PM2.5-induced gene signatures: A Machine Learning Approach to Individual Exposure Prediction

PM2.5, a prominent air pollutant, has been extensively studied through gene expression profiling, demonstrating its activity to modulate gene expressions linked to adverse health outcomes and diseases. Despite the studies, population-based cohort investigations exploring the relationship between PM2.5 exposure and specific gene sets remain limited. In this study, we conducted an unbiased transcriptomic profiling examined gene expression in a PM2.5-exposed mouse model to identify PM2.5-responsive genes. The genes were validated in human cell lines and population-based cohort investigation. Two cohorts comprising healthy older adults aged 65 and above, residing in the regions with contrasting PM2.5 levels, were used to assess the expression of the genes. These gene expression profiles were further used to develop predictive models for PM2.5 exposure by employing logistic regression and decision tree algorithms. Our results showed that five genes (FAM102B, PPP2R1B, OXR1, ITGAM, and PRP38B) increased with PM2.5 exposure in both cell-based assays and population-based cohort studies. The predictive models exhibited high accuracy in assigning high-low PM2.5 exposure, enabling the integration of gene biomarkers into public health practices. This approach may transform air pollution management, shifting from reliance solely on environmental monitoring to incorporating biological and genetic monitoring that reflects individual-level health risks.