Machine learning assisted discovery of synergistic interactions between environmental pesticides, phthalates, phenols, and trace elements in child neurodevelopment

Abstract

A growing body of literature suggests that higher developmental exposure to individual or mixtures of environmental chemicals (ECs) is associated with autism spectrum disorder (ASD). However, the effect of interactions among these ECs is challenging to study. We introduced a composition of the classical exposure-mixture Weighted Quantile Sum (WQS) regression, and a machine-learning method called signed iterative random forest (SiRF) to discover synergistic interactions between ECs that are (1) associated with higher odds of ASD diagnosis, (2) mimic toxicological interactions, and (3) are present only in a subset of the sample whose chemical concentrations are higher than certain thresholds. In the case-control Childhood Autism Risks from Genetics and Environment study, we evaluated multi-ordered synergistic interactions among 62 ECs measured in the urine samples of 479 children in association with increased odds for ASD diagnosis (yes vs. no). WQS-SiRF discovered two synergistic two-ordered interactions between (1) trace-element cadmium(Cd) and alkyl-phosphate pesticide - diethyl-phosphate(DEP); and (2) 2,4,6-trichlorophenol(TCP-246) and DEP metabolites. Both interactions were suggestively associated with increased odds of ASD diagnosis in a subset of children with urinary concentrations of Cd, DEP, and TCP-246 above the 75th percentile. This study demonstrates a novel method that combines the inferential power of WQS and the predictive accuracy of machine-learning algorithms to discover interpretable EC interactions associated with ASD.

Detecting Shape-based Interactions among Environmental Chemicals using an Ensemble of Exposure-Mixture Regression and Interpretable Machine Learning Tools

“…We excluded product or projection-based algorithms that do not directly represent the collective activity of the chemical exposures. Note that improved variations of the SiRF algorithm using a repeated hold-out stage (Midya et al (2023a) and Midya et al (2023b)) exist in the literature. But we simply focus on the original SiRF algorithm (Basu et al (2018)) without any repeated hold-out.…”

Prenatal Lead Exposure Is Associated with Reduced Abundance of Beneficial Gut Microbial Cliques in Late Childhood: An Investigation Using Microbial Co-Occurrence Analysis (MiCA)

“…23−25 Tree-based machine learning models can provide a natural and computationally efficient solution to such construction. 20 Even with a substantial number of taxa, these models can create multiple threshold-based combinations of taxa predictive of the outcome of interest. Still, the challenge remains in interpretability since most machine learning models are generally black-box, creating a tension between prediction quality and meaningful interpretability.…”

Prenatal Pb exposure is associated with reduced abundance of beneficial gut microbial cliques in late childhood: an investigation using Microbial Co-occurrence Analysis (MiCA)

“…However, finding microbial clique associated with an outcome of interest is challenging because of (1) considerable computational complexity as the number of taxa increases and (2) limitations of sample size in most human microbiome studies. Multiple methods exist where multi-ordered microbial cliques can be pre-specified or hard-coded in the models; however, such strategies might ignore many plausible and informative combinations and could be underpowered due to restrictions on sample size (Gibson 2021;Joubert et al 2022;Vishal Midya et al 2023). On the other hand, microbial cliques can be potentially discovered using projection-based dimensionality reduction techniques.…”

“…Such threshold-based construction carries considerable similarity with toxicological threshold-based interactions (Gennings et al 1997;Hamm et al 2005;Yeatts et al 2010). Tree-based machine-learning models can provide a natural and computationally efficient solution to such construction (Vishal Midya et al 2023). Even with a substantial number of taxa, these models can create multiple threshold-based combinations of taxa predictive of the outcome of interest.…”