Population Cohort-Validated PM(2.5)-Induced Gene Signatures: A Machine Learning Approach to Individual Exposure Prediction.

Transcriptomic profiling has shown that exposure to PM(2.5), a common air pollutant, can modulate gene expression, which has been linked to negative health effects and diseases. However, there are few population-based cohort studies on the association between PM(2.5) exposure and specific gene set expression. In this study, we used an unbiased transcriptomic profiling approach to examine gene expression in a mouse model exposed to PM(2.5) and to identify PM(2.5)-responsive genes. The gene expressions were further validated in both the human cell lines and a population-based cohort study. Two cohorts of healthy older adults (aged ≥ 65 years) were recruited from regions characterized by differing levels of PM(2.5). Logistic regression and decision tree algorithms were then utilized to construct predictive models for PM(2.5) exposure based on these gene expression profiles. Our results indicated that the expression of five genes (FAM102B, PPP2R1B, OXR1, ITGAM, and PRP38B) increased with PM(2.5) exposure in both cell-based assay and population-based cohort studies. Furthermore, the predictive models demonstrated high accuracy in classifying high-and-low PM(2.5) exposure, potentially supporting the integration of gene biomarkers into public health practices.