With the large, and ever increasing, number of poly- and perfluorinated compounds present in the environment, the task of predicting their movement, accumulation, and reactivity for the purposes of capture and/or remediation becomes an ever more daunting task. With over 12,000 compounds in the EPA’s PFAS Master List alone, the notion of fully characterizing each of these compounds experimentally is impractical. Computational methodologies provide a route to predict the properties of these compounds. Some methods rooted in quantum chemistry can attain higher accuracy, albeit at a high computational cost. Less expensive methods may use empirical approaches for higher speed, but also tend to result in reduced accuracy. Both of these points are exacerbated by the need to continually apply the same methodology as additional PFAS molecules continue to be tabulated. Data-driven approaches, such as machine learning, can potentially provide an extensible solution to this problem.  In this effort, we aim to use machine learning methods to reproduce environmentally relevant properties of poly- and perfluorinated compounds from available experimental and computational data. Beginning from text-based (or similar) representations, we apply multiple techniques in the hope of arriving at a robust model that can provide both a rapid and reasonable first approximation of these properties.

Using chemical identifiers to predict environmentally relevant properties of poly- and perfluorinated compounds

Where thousands of chemistry professionals meet to share ideas and advance scientific and technical knowledge. Spring 2023 abstracts, posters, and presentations will launch at the end of March.

ACS Spring 2023

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.