Abstract
AIM: In this research, we leveraged bioinformatics and machine learning to pinpoint key risk genes associated with occupational benzene exposure and to construct genomic and algorithm-based predictive risk assessment models. SUBJECT AND METHODS: We sourced GSE9569 and GSE21862 microarray data from the Gene Expression Omnibus. Utilizing R software, we performed an initial screen for differentially expressed genes (DEGs), which was followed by the enrichment analyses to elucidate the affected functions and pathways. Subsequent steps included the application of three machine learning algorithms for key gene identification, and the validation of these genes within both a cohort exposed to benzene and a benzene-exposed mice model. We then conducted a functional prediction analysis on these genes using four machine learning models, complemented by GSVA enrichment analysis. RESULTS: Out of the data, 40 DEGs were identified, primarily linked to cytokine signaling, lipopolysaccharide response, and chemokine pathways. NFKB1, PHACTR1, PTGS2, and PTX3 were pinpointed as significant through machine learning. Validation confirmed substantial changes in NFKB1 and PTX3 following exposure, with PTX3 emerging as paramount, suggesting its utility as a diagnostic biomarker for benzene damage. CONCLUSION: Risk assessment models, informed by oxidative stress markers, successfully discriminated between benzene-injured patients and controls.