Composite pollutants have become a major environmental issue threatening public health due to their characteristics of long-term persistence, bioaccumulation and high toxicity. These pollutants not only exhibit complex high-dimensional co-exposure patterns and interactions but may also influence the onset and progression of diseases through nonlinear dose-response relationships. The primary challenge in current research is that traditional epidemiological methods focus on direct associations between exposure and outcomes, making it difficult to reveal the synergistic/antagonistic effects of composite pollutants and their perturbation mechanisms on disease pathways. Although the rapid development of metabolomics has provided new perspectives for elucidating related biological pathways, the existing mediation analysis face notable limitations in high-dimensional data modeling and causal inference.
Therefore, this project proposes to integrate machine learning, Bayesian statistics, and causal inference methods to construct a multi-level and highly robust analytical framework for systematically elucidating the complex pathway mechanisms linking “composite pollutant exposure-metabolomics-disease phenotypes.” Theoretically, the well-established Bayesian statistical framework provides robust support for modeling high-dimensional data. Then rapidly evolving causal inference methods effectively dissect complex biological pathways. Furthermore, the availability of large-scale datasets such as UK Biobank, which includes high-resolution metabolomics profiles across diverse populations, provides an unprecedented opportunity to decode the metabolic pathways and their role in disease outcomes. The ultimate goal is to develop a statistical methodological system for Bayesian high-dimensional causal mediation joint analysis, addressing theoretical gaps in the field.
