A Gaussian process approach facilitates the identification of robust biomarkers for exposure to complex pesticide mixtures.

Biomarkers can provide a high-throughput and accurate assessment of the impact of complex chemical mixtures in the environment on organisms, but their identification through gene expression analysis is hindered by noise, synergistic interactions, and non-linear expression patterns. We generated finely resolved transcriptomic data from the ecotoxicological model species Folsomia candida exposed to two binary pesticide mixtures: One combining two neonicotinoid insecticides (imidacloprid and clothianidin) and the other a neonicotinoid (imidacloprid) with an azole fungicide (cyproconazole). Using these datasets, we developed a Gaussian Process (GP) framework to identify robust gene expression biomarkers, accounting for non-linear and synergistic interaction effects across experiments. Joint analysis of two binary mixtures increased the overlap of differentially expressed genes (DEGs) compared to separate analyses, improving robustness. In simulations, GP models outperformed linear models, accurately fitting complex, non-linear concentration-response relationships. Four biomarkers, three for neonicotinoids (ARRD, SMCT and nAchR) and one for azole fungicides (CYP), identified through this framework, were empirically validated and confirmed to be specifically responsive to their target pesticide, even under co-exposure. These findings highlight the effectiveness of GP models for mixture exposure transcriptomics and their broader applicability to other omics data and research fields.