Advancing AI/ML-driven chemical exposomics to identify biologically relevant environmental exposures

Exposomics aims to systematically measure environmental exposures that an individual experiences across all stages of life, from conception to death, and link them to health and disease states. The ability of high-resolution mass spectrometry (HRMS) to perform non-targeted detection of thousands of chemicals in biological and environmental matrices has made this goal feasible. Artificial intelligence and machine learning (AI/ML) have become a transformative force in parsing this complexity, primarily by enhancing the ability to identify previously unknown chemicals from spectral data. However, this perspective argues that while expanding the catalog of detected chemicals is essential, the ultimate goal of AI/ML in chemical exposomics is shifting towards functional studies of chemicals with the greatest potential to disrupt biological systems and cause disease, referred to as functional chemical exposomics, which is the comprehensive study of biological activities, mechanistic pathways, and phenotypic responses triggered by all environmental exposures throughout a lifetime. We propose a paradigm shift in which AI transitions from a "discovery engine" to a "functional prediction engine," integrating multimodal data (chemical structures, predicted toxicity, and biological response profiles) for exposomic screening. This article outlines the framework of this vision, discusses pioneering research supporting this transition, and explores key challenges and future directions for achieving proactive, health-related functional chemical exposomics.