Due to their pervasiveness and persistence in the environment, per- and polyfluoroalkyl substances (PFAS) are gaining interest as an environmental and biological concern. Applications of PFAS compounds include a multitude of different PFAS species, which hinders the development of an experimental knowledgebase of informative properties, characterizations, and degradation mechanisms by limiting studies to only a small number of systems. To help mitigate this issue, we propose to develop an advanced database of PFAS properties from heterogeneous data sources on the web – building a platform that identifies existing and new PFAS compounds from new research in a wide variety of data formats, including online databases and repositories (e.g., text-mining approaches on patents and publications in archives). The collected data will be curated and maintained in a single curated materials data lake (CMDL) for PFAS. Furthermore, we will address data gaps of the chemical properties using advanced computational chemistry approaches. Novel featurizations of PFAS for machine learning (ML) algorithms will be developed by combining gathered and predicted data. We will also include key structural information and properties relevant to fate and transport and species reactive towards C-F bonds. We will develop a formal approach for capturing the salient details of the atomic structure of PFAS using digital representations of atomic structure that will allow the computationally efficient feature engineering needed to employ machine learning toolsets. Such ML models may be used to predict dissociation energy for new PFAS compounds by learning continuously from the aggregated data. Thus, this work lays the foundation for our future efforts: developing AI/ML-guided approaches for PFAS remediation. These models can then be used to discover new solutions with high potential for defluorination.

Advanced featurizations for AI/ML-guided PFAS remediation

ACS Spring Meeting 2021

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