Identifying Endocrine-Disrupting Plastic Additives using Machine Learning

About This Grant

ABSTRACT Plastic additives are widely used in consumer products, yet thousands of plastic additives remain uncharacterized for their potential to disrupt endocrine function - posing significant public health risks. This project aims to develop an integrated computational (Aim 1) and experimental (Aim 2) workflow to systematically predict and validate the endocrine-disrupting potential of plastic additives. In Aim 1, we will design novel machine learning models trained on publicly available datasets to predict AR and ERα modulating activity of plastic additives and then used to predict the potential effects of all plastic additives to select the most promising based on novelty and predictive uncertainty for further in vitro and in vivo testing. In Aim 2, we will validate our predictions through a multi-step experimental characterization approach using our in-house AR and ERα assays, followed by dose-response studies in AR- and ERα-responsive cell lines to measure target gene activation and cell proliferation. The top three plastic additives with the strongest in vitro effects will be further evaluated in vivo using mice to assess systemic hormonal changes caused by the plastic additives. This work will have a substantial positive societal impact by establishing a first-in-kind machine learning-assisted predictive toxicological model to pinpoint plastic additives of highest concern to induce adverse health effects as well as generate a large dataset of plastic additive effects on endocrine function. Taken together, this work can serve to provide policy guidance on plastic additives to ban or remove from products, with potentially beneficial health outcomes for billions of consumers.