Methodology for Good Machine Learning with Multi-Omics Data

Screening structure and predicting toxicity of pesticide adjuvants using molecular dynamics simulation and machine learning for minimizing environmental impacts

Surfactants as synergistic agents are necessary to improve the stability and utilization of pesticides, while their use is often accompanied by unexpected release into the environment. However, there are no efficient strategies available for screening low-toxicity surfactants, and traditional toxicity studies rely on extensive experimentation which are not predictive. Herein, a commonly used agricultural adjuvant Triton X (TX) series was selected to study the function of amphipathic structure to their toxicity in zebrafish. Molecular dynamics (MD) simulations, transcriptomics, metabolomics and machine learning (ML) were used to study the toxic effects and predict the toxicity of various TX. The results showed that TX with a relatively short hydrophilic chain was highly toxic to zebrafish with LC50 of 1.526 mg/L. However, TX with a longer hydrophilic chain was more likely to damage the heart, liver and gonads of zebrafish through the arachidonic acid metabolic network, suggesting that the effect of surfactants on membrane permeability is the key to determine toxic results. Moreover, biomarkers were screened through machine learning, and other hydrophilic chain lengths were predicted to affect zebrafish heart health potentially. Our study provides an advanced adjuvants screening method to improve the bioavailability of pesticides while reducing environmental impacts.