White 411 or Zoom Webinar ID: 940 7438 8634 Passcode: 357363

Machine Learning Modeling for the Reactivity of Contaminants in Engineered and Natural Environments

Abstract: To mitigate and assess the risks associated with thousands of contaminants in engineered and natural environments, we need accurate predictive models that can readily provide reasonable estimates of their reactivity, both during important water treatment processes, including advanced oxidation processes (AOPs) and adsorption, and in the environment, such as through sorption on soils/sediments. However, conventional models rely heavily on quantitative structure-activity relationships (QSARs) between molecular descriptors and chemical activity which have multiple limitations, such as small numbers and narrow-scopes of contaminants involved, tedious calculations of molecular descriptors, and ignorance of adsorbent properties. In this talk, we’ll discuss our recent progresses in using machine learning algorithms to develop more powerful, robust, and trustworthy predictive models. Specifically, we have 1) mined the literature and available databases to obtain large datasets of contaminant reactivity in AOPs and adsorption; 2) experimentally quantified the activity of selected contaminants in AOPs or adsorption; 3) developed predictive machine learning models for the activity of contaminants based on the data from the above two objectives; and 4) interpreted the obtained machine learning models to make them trustable and defined their applicability domains.