Per- and Polyfluoroalkyl Substances (PFAS) are persistent in our environment due to their strong carbon–fluorine (C–F) bonds causing a pervasive exposure of humans to PFAS through food and drinking water. Treatment/removal of the PFAS pollutant requires identification and classification of thousands of varieties of PFAS. The adoption of appropriate molecular descriptors is crucial in establishing structure-activity/structure-property linkages which assist in the treatment/removal of PFAs. There exist various molecule descriptors which capture distinct aspects of chemical structures. However, many of these descriptors are inter-dependent and sometimes capture the same information. Therefore, the selection of the appropriate molecular descriptors to perform rigorous analysis is a nontrivial task. Recent works have shown that the incorporation of relatively simple descriptions of the molecular structure leads to successful predictions of bond dissociation energy using a variety of AI/ML algorithms. In this work, a rigorous methodology for the characterization of the molecular structure has been explored using voxelization techniques. The explored voxelization techniques were used for the predictions of bond dissociation energy (BDE). One of the main hurdles in the development and testing of powerful molecular descriptors is the lack of an up-to-date database which sufficiently captures the electrical, physical, structural, and chemical properties which can be used to characterize PFAS. All of this information is not readily available from a single source. However, such information can be aggregated from the combination of information available from public databases, in-house datasets and peer reviewed journals. The Multiscale Technologies platform was used to compile this disparate information into a comprehensive dataset, and facilitate its curation, feature engineering, and deployment in ML models. Specifically, data was aggregated from the publicly available EPA database as well recent publications in order to develop and extend a ML model capable of predicting BDE for a wide variety of PFAS.

Curation and enrichment of per- and polyfluoroalkyl substances databases using feature engineering approaches for AI/ML

Fall 2022 abstracts, posters, and presentations will launch in mid-September.

ACS Fall 2022

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