A statistical workflow for analyzing the untargeted chemical exposome and metabolome in epidemiologic studies using high-dimensional mixture methods

Humans are exposed to upwards of thousands of chemicals simultaneously, but research has traditionally focused on the health effects of only one chemical at a time. Single-chemical analyses not only underestimate total health risk, but also ignore bias from multicollinearity and co-exposure confounding between chemicals. Advanced statistical mixture methods address these challenges and allow us to both estimate the cumulative health effect of chemical mixtures and identify the strongest chemical contributors. At the same time, untargeted chemical exposome profiling through high-resolution mass spectrometry (HRMS) now supports measurement of over 100,000 chemical signals in biospecimens. However, most mixture methods cannot evaluate untargeted exposome data containing more chemical variables than samples. Weighted quantile sum regression with its recent random subsets implementation (WQS_RS) is a unique, statistically powerful mixture method for high-dimensional exposome data. It estimates weights of chemicals towards the mixture index over many different repetitions in which only a small random subset of chemicals is used at a time, thus de-correlating data and avoiding overfitting. In this paper, we discuss our statistical workflow and important considerations for the application of WQS_RS to exposome epidemiology, including manual quantization for non-detects, custom repeated holdouts for matched data, pre-selection of exogenous chemicals, parameter decisions, interpretation options, and visualizations. We then describe its application to functional pathway enrichment analysis with integrated exposome-metabolome data to explore underlying biological mechanisms. These data science approaches will enable exposome epidemiology to discover previously unknown risk factors, estimate cumulative health risk from total chemical mixtures, and gain mechanistic insight.