Molecular Screening and Toxicity Estimation of 260,000 Perfluoroalkyl and Polyfluoroalkyl Substances (PFASs) through Machine Learning

Abstract

Perfluoroalkyl and polyfluoroalkyl substances (PFASs) are a class of chemicals widely used in industrial applications due to their exceptional properties and stability. However, they do not readily degrade in the environment and are linked to contamination and adverse health effects in humans and wildlife. To find alternatives for the most commonly used PFAS molecules that maintain their desirable chemical properties but are not adverse to biological lifeforms, a novel approach based upon machine learning is utilized. The machine learning model is trained on an existing set of PFAS molecules to generate over 260,000 novel PFAS molecules, which we dub PFAS-AI-Gen. Using molecular descriptors with known relationships to toxicity and industrial suitability followed by molecular docking and molecular dynamics simulations, this set of molecules is screened. In this manner, increasingly complex calculations are performed only for candidate molecules that are most likely to yield the desired properties of low binding affinity toward two selected protein receptors, the human pregnane x receptor (hPXR) and peroxisome proliferator-activated receptor γ (PPAR-γ), and high industrial suitability, defined by critical micelle concentration (CMC). The selection criteria of low binding affinity and high industrial suitability are relative to the popular PFAS alternative GenX. hPXR and PPAR-γ are selected as they are PFAS targets and facilitate a variety of functions, such as drug metabolism and glucose regulation, respectively. Through this approach, 22 promising new PFAS substitutes that may warrant experimental investigation are identified. This integrated approach of molecular screening and toxicity estimation may be applicable to other chemical classes.

High-Level Quantum Chemical Prediction of C–F Bond Dissociation Energies of Perfluoroalkyl Substances

“…Notably, DFT-calculated BDEs of over 100 PFASs (564 distinct C−F BDEs) were used to predict C−F BDEs and identify structure−BDE relationships, 14 and patterns in computed protein binding affinities were used to propose new PFASs for industrial use. 15 However, the performance of both methods is also known to be highly system-dependent, and its accuracy is usually estimated through statistical benchmarking against large test sets of high-quality theoretical or experimental data.…”

“…Computational modeling is often used to elucidate structure–reactivity relationships in PFASs through calculation of carbon–fluorine bond dissociation energies (BDEs), as nonfluorinated products observed in these reductive degradation experiments suggest that C–F bond cleavage is a key step in the mechanism. − Due to their computational efficiency, density functional theory (DFT) and molecular mechanics (MM) methods are often employed to explain the structure activity relationships of different classes of PFASs. ,− The collection of these computed properties forms the basis for many machine learning applications. Notably, DFT-calculated BDEs of over 100 PFASs (564 distinct C–F BDEs) were used to predict C–F BDEs and identify structure–BDE relationships, and patterns in computed protein binding affinities were used to propose new PFASs for industrial use . However, the performance of both methods is also known to be highly system-dependent, and its accuracy is usually estimated through statistical benchmarking against large test sets of high-quality theoretical or experimental data.…”

Building Chemical Intuition about Physicochemical Properties of C8-Per-/Polyfluoroalkyl Carboxylic Acids through Computational Means

“…Physicochemical properties have been predicted within chemical accuracy by DFT, COnductor-like Screening MOdel for Realistic Solvents (COSMO-RS), SPARC, Open structure–activity/Property Relationship App (OPERA), EPI Suite, and machine-learning methods for a variety of compounds, including PFAS. ,− Herein, we elect to utilize the COSMO-RS model to predict relevant physicochemical properties. Quantum mechanical calculations paired with COSMO-RS models present a reliable method to predict physicochemical properties. ,− This method is especially relevant for PFAS as fluorine chemistry exhibits hyperconjugative effects and novel chemical properties that may often conflict with common chemical intuition .…”

Binding of Per- and Polyfluoroalkyl Substances (PFAS) to the PPARγ/RXRα–DNA Complex

“…Because of the large number of PFAS compounds, recently, machine learning (ML) approaches have also been utilized to predict the binding between PFAS and nuclear receptors. 78 One of the most recent approaches considered the binding of 4464 PFAS to PPARα and γ and the thyroid hormone receptor. 79 The authors concluded that the binding energies of PFAS to thyroid hormone receptors are 2−3 kcal mol −1 stronger than to PPARγ.…”

“…As well, a machine learning strategy was utilized to identify novel PFAS compounds that may be less toxic than current PFAS such as GenX. 8,9,78 Herein, a variety of PFAS were investigated to consider their effect on the activity of the PPARγ/RXRα-DNA complex. In total, nine PFAS with different chain lengths and functional groups were selected, along with L-carnitine.…”