An Enhanced Protocol to Expand Human Exposome and Machine Learning-Based Prediction for Methodology Application

The human exposome remains limited due to the challenging analytical strategies used to reveal low-level endocrine-disrupting chemicals (EDCs) and their metabolites in serum and urine. This limits the integrity of the EDC exposure assessment and hinders understanding of their cumulative health effects. In this study, we propose an enhanced protocol based on multi-solid-phase extraction (multi-SPE) to expand human exposome with polar EDCs and metabolites and train a machine learning (ML) model for methodology prediction based on molecular descriptors. The protocol enhanced the measurement of 70 (25%) and 34 (12%) out of 295 well-acknowledged EDCs in serum and urine compared to the hydrophilic-lipophilic balance sorbent alone. In a nontarget analysis of serum and urine from 20 women of childbearing age in a cohort of 498, controlling occupational factors and daily behaviors for high chemical exposure potential, the multi-SPE protocol increased the measurement of 10 (40%) and 16 (53%) target EDCs and identification of 17 (77%) and 70 (36%) nontarget chemicals (confidence >= level 3) in serum and urine, respectively. Interestingly, the ML model predicted that the multi-SPE protocol could identify an additional 38% of the most bioactive chemicals. In conclusion, the multi-SPE protocol advances human exposome by expanding the measurement and identification of exposure profiles.