Computational insights into exploring the potential effects of environmental contaminants on human health

With rapid industrialization and urbanization, the increasing prevalence of air and water pollutants poses a significant threat to public health. Traditional research methods, such as epidemiological studies and in vitro/in vivo experiments, provide valuable biological insights but are often costly, time-consuming, and limited in scale. To address this gap, this study develops a machine learning-based approach to predict the carcinogenicity of pollutants. Using the dataset of carcinogenic and non-carcinogenic molecules that we collected, the pretrained KPGT model trained with molecular fingerprints and descriptors achieved an AUC of 0.83, surpassing traditional machine learning models. To validate this model, common pollutants from air and water sources were analyzed. Further clustering classified these pollutants into five distinct groups. Target prediction analysis identified key genes associated with representative pollutant molecules, such as MAPK1, MTOR, and PTPN11. GO and KEGG pathway analyses, along with survival analysis, revealed potential carcinogenic mechanisms and prognostic implications. Our findings contribute to improved pollution risk assessment and evidence-based environmental policy development, ultimately aiding in the mitigation of pollutant-related health risks.