2025 AIChE Annual Meeting

(588at) Screening of Solvents for Capturing PFAS Via a Data Science Approach

The development and broad application of single-layer and multilayer (ML) plastics in the food packaging, pharmaceuticals, and cosmetics industries have served various purposes, including ensuring consumer product safety, shelf life, and handling (Barry, 2022). However, the recycling rate for these materials is low, particularly in the US, which only reaches 9% (Geyer et al. 2017). The emergence of solvent-based-physical plastic separation processes, such as the Solvent Targeted Recovery And Precipitation (STRAP™) process, has facilitated the recycling of these plastic wastes that cannot be mechanically recycled, resulting in the heightened recyclability of post-industrial waste plastics (Walker et al. 2020; Zhao et al. 2018). This technology involves the sequential separation of ML plastics into their various polymer constituents over a series of solvent washes, recovering high-quality polymer resins for reuse in food packaging and other consumer products. However, with the emergence of the Chemicals of Concern (Aurisano et al. 2021; Zimmermann et al. 2022) such as Per-and Polyfluoroalkyl Substances (PFAS) that may be non-intentionally introduced (Tumu et al. 2024) into post-consumer waste (PCW) plastics somewhere along the plastic waste management supply chain such as the material recovery facilities, human use, or the environment, current mechanical and solvent-based recycling infrastructures face a significant challenge in ensuring contaminant-free recycled plastics. Moreover, the ubiquitousness of PFAS in the environment and their endocrine-disruptive nature has prompted legislative bans on the industrial use of fluorinated compounds as processing aids, including fluoropolymers across the US, particularly in the food packaging industry, to varying detection limits (Bulawska et al. 2025; Norris et al. 2022; Zheng et al. 2023). Therefore, extending solvent-based plastic recycling infrastructures to remediate short-chain PFAS from PCW plastics is paramount for increasing the recyclability of plastic wastes. However, unlike the finite number of polymers, 15,000 EPA-identified small-molecule PFAS could contaminate PCW plastics (Gaines et al. 2023). Therefore, this study applies data science approaches to quickly identify solvents to capture these PFAS from plastic wastes – ensuring PFAS-free post-consumer recycled plastics.

This study expands the STRAP process to incorporate the capture of PFAS using solvents by treating them as an additional layer in selective dissolution. The objective is to identify solvents that selectively dissolve PFAS while preserving other polymers present in the PCW plastics. This approach leverages the partition coefficient (LogD) and the solubility of PFAS in various solvents—predicted via molecular-scale modeling—to identify solvents that facilitate the preferential partitioning of PFAS (LogD > 0) from the polymer matrix into the solvent phase. Specifically, the partition coefficients of each PFAS in different polymers and solvents are essential for evaluating a solvent’s ability to extract PFAS from a given polymer. Additionally, the solubility of each PFAS in different solvents must be assessed to determine the effectiveness of each solvent in capturing PFAS. This molecular-scale modeling involves performing density functional theorem (DFT) calculations on each PFAS to obtain their surface screening charges (i.e., sigma profiles) that necessitate the prediction of relevant molecular properties and the use of the COnductor-like Screening MOdel for Real Solvents (COSMO-RS) software, grounded in quantum chemistry principles, that enables accurate prediction of their LogD and solubilities. However, predicting the LogD of the 15,000 PFAS in 10 different polymers present in PCW plastics in 1000 solvent requires ~15,000,000 molecular simulations or experiments and over 15,000 human hours of simulation time, eliminating the feasibility of experimental screening approaches and requiring significant computational resources. Therefore, PFAS remediation from PCW plastics requires leveraging data science tools for a more efficient PFAS capture strategy. Hypothesizing that structurally similar PFAS have similar molecular properties, this work identifies families of structurally similar PFAS via dimensionality reduction approaches and identifies solvents that can target each family. This approach significantly reduces the number of molecular simulations by 100,000 times and the computational time required.

The Molecular ACCess System (MACCS) keys, a one-hot encoding molecular representation, obtained from the open-source rdkit python library, were adopted to represent each PFAS molecule, and t-SNE, a dimensionality reduction technique, was used to transform the structural data from a high-dimensional space to a low-dimensional space for easy visualization. This analysis successfully identified seven distinct families of PFAS having different functional groups and structural features. To test our hypothesis founded on structural similarity, we selected 50 PFAS spanning the entire latent space to perform a pair-wise comparison between their similarity score and LogD magnitude difference. Using polyethylene (PE) as the reference polymer, the LogD values of each PFAS were predicted in five different solvents using the COSMO-RS software. Our results show that 70% similarity is the threshold value for our hypothesis to hold. Therefore, using this threshold value, we performed a guided clustering to determine the number of clusters that would ensure a minimum of 70% structural similarity within each cluster across the entire PFAS latent space. Due to the inherent inability of existing dimensionality techniques, such as t-SNE and UMAP, to perform such guided clustering, a graph-based clustering approach was adopted. Given the 70% threshold value, an undirected graph was developed to represent the interconnectivity between PFAS based on this similarity threshold value (i.e., nodes representing PFAS and edges with at least 70% similarity). While ongoing works involve examining the topology of our undirected graph to identify clusters of interconnected PFAS, our results show that 800 PFAS are structurally unique, having no connection with other PFAS. This result provides insights into PFAS that may be difficult or easy to target via solvent-based approaches. Our results also show that tetrahydrofuran (THF), a biomass-based solvent, proves promising in targeting a wide variety of PFAS from PE. Other ongoing works include applying our solvent screening framework to an existing ML plastic film to identify families of PFAS targeted using selected solvents proposing solvent-based plastic recycling also as a possible pretreatment step for other plastic recycling technologies.

References:

Aurisano, N., Huang, L., i Canals, L. M., Jolliet, O., & Fantke, P. (2021). Chemicals of concern in plastic toys. Environment International, 146, 106194.

Barry, M. A. (2022). The Science and Technology of Flexible Packaging: Multilayer Films from Resin and Process to End Use. Elsevier.

Bulawska, N., Sosnowska, A., Kowalska, D., Stępnik, M., & Puzyn, T. (2025). PFAS (per-and polyfluorinated alkyl substances) as EDCs (endocrine-disrupting chemicals)-Identification of compounds with high potential to bind to selected terpenoids NHRs (nuclear hormone receptors). Chemosphere, 370, 143967.

Gaines, L. G., Sinclair, G., & Williams, A. J. (2023). A proposed approach to defining per‐and polyfluoroalkyl substances (PFAS) based on molecular structure and formula. Integrated Environmental Assessment and Management, 19(5), 1333-1347.

Geyer, R., Jambeck, J. R., & Law, K. L. (2017). Production, use, and fate of all plastics ever made. Science advances, 3(7), e1700782.

Norris, T. H., Preheim, E. J., Prero, J., & Wagner, W. (2022). Legal and Litigation Updates on PFAS in Products. FDLI Update, 7.

Tumu, K., Vorst, K., & Curtzwiler, G. (2024). Understanding intentionally and non-intentionally added substances and associated threshold of toxicological concern in post-consumer polyolefin for use as food packaging materials. Heliyon, 10(1).

Walker, T. W., Frelka, N., Shen, Z., Chew, A. K., Banick, J., Grey, S., ... & Huber, G. W. (2020). Recycling of multilayer plastic packaging materials by solvent-targeted recovery and precipitation. Science advances, 6(47), eaba7599.

Zhao, Y. B., Lv, X. D., & Ni, H. G. (2018). Solvent-based separation and recycling of waste plastics: A review. Chemosphere, 209, 707–720.

Zheng, G., Eick, S. M., & Salamova, A. (2023). Elevated levels of ultrashort-and short-chain perfluoroalkyl acids in US homes and people. Environmental Science & Technology, 57(42), 15782-15793.

Zimmermann, L., Scheringer, M., Geueke, B., Boucher, J. M., Parkinson, L. V., Groh, K. J., & Muncke, J. (2022). Implementing the EU Chemicals Strategy for Sustainability: The case of food contact chemicals of concern. Journal of Hazardous Materials, 437, 129167.