Per- and polyfluoroalkyl substances (PFASs) concentrations in some contaminated, public ground and surface waters are orders of magnitude higher than EPA limits for drinking water and are potentially toxic to living organisms. Thus, there is an urgent need for developing novel technologies for mitigating PFAS contaminations. Computational advancements offer an efficient and cost-effective avenue for developing new alternatives. Here, we combine bioinformatics and machine learning approaches to mine for and identify the types of environmental microorganisms that harbor catabolic enzymes of interest. We initially focus our efforts on microorganisms such as Acidimicrobium and Delftia acidovorans and their pathways/enzymes that degrade PFAS. We then use these as queries via Hidden Markov Modeling (HMM) to identify novel enzyme variants and other diverse microorganisms with the potential for PFAS degradation from publically-available sequence repositories (e.g., metagenomes derived from organohalogen-contaminated sites). Using a combination of multiple sequence alignments, phylogenies, sequence similarity networks, and structural prediction analyses we will select promising enzyme candidates for modeling, expression, and testing. We implement a machine learning model to create learned representations of both partial and complete protein sequences, allowing these sequences to be directly input to additional machine learning analysis methods without the need for a priori knowledge. We develop additional supervised machine learning models to predict the classes of PFAS-degrading microorganisms and enzymes to use for optimized PFAS degradation. We believe that bioinformatics and AI/ML algorithms will tremendously aid in the advancement of the biological degradation of PFASs.

Combined bioinformatics and machine learning approach for biological degradation of PFASs

ACS Spring Meeting 2021

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