(515e) Physics-Informed Neural Networks for Accelerating PFAS Sorbent Discovery: A Hybrid Mechanistic–Data Driven Approach

Per- and polyfluoroalkyl substances (PFAS), known as “forever chemicals”, persist in ecosystems and pose severe threats to human health due to their bioaccumulative nature and resistance to degradation [1]. Sorbent-based technologies, encompassing both adsorption (surface adhesion) and absorption (bulk uptake) are critical for PFAS sequestration. However, the discovery of high-performance sorbents is hindered by the complex, multifactorial nature of solute-sorbent interactions [2], which vary across material classes (e.g., activated carbon, hydrogels, metal-organic frameworks) and PFAS chemistries (e.g., ionic vs. non-ionic species). Traditional approaches, which treat adsorption and absorption as separate phenomena or rely on empirical trial-and-error, lack the efficiency and mechanistic rigor needed to optimize sorbent designs for real-world conditions [3, 4].

To address this gap, we present a Hybrid Mechanistic–Data-Driven Framework leveraging Physics-Informed Neural Networks (PINNs) to unify adsorption and absorption mechanisms within a single predictive model. Our approach integrates first-principles physics—such as surface adsorption kinetics (e.g., Langmuir isotherms, electrostatic interactions) and bulk absorption dynamics (e.g., diffusion coefficients, partition equilibria)—with machine learning trained on literature data. This dataset spans sorbent properties (surface area, porosity, hydrophobicity), PFAS characteristics (molecular weight, functional groups, solubility), and operational parameters (pH, temperature, co-contaminant effects).

The framework advances sorbent discovery through three steps:

1. Dual-Mechanism Feature Engineering: Physicochemical descriptors differentiate adsorption-dominated processes (e.g., van der Waals interactions on activated carbon) from absorption-driven uptake (e.g., PFAS diffusion into hydrogel matrices).
2. Physics-Constrained PINN Architecture: Governing equations for adsorption (e.g., pseudo-second-order kinetics) and absorption (e.g., Fickian diffusion) are embedded into loss functions, ensuring predictions align with domain knowledge.
3. Mechanistic Interpretability Tools: Shapley value analysis and sensitivity testing quantify the relative contributions of adsorption vs. absorption mechanisms for diverse sorbent-PFAS pairs.

Validation against literature datasets demonstrates the framework’s versatility. For adsorption-dominant systems (e.g., amine-functionalized MOFs), the model predicts PFAS uptake with >70% accuracy by prioritizing surface charge and pore size. For absorption-dominant systems (e.g., cross-linked polymer hydrogels), it identifies diffusion-limited uptake, guiding optimization of cross-linking density and hydrophobicity.

The hybrid PINN reduces the experimental screening burden by 50–70% compared to conventional methods while maintaining interpretability. For example, it revealed that alkyl chain length in PFAS dictates absorption efficiency in hydrophobic polymers, whereas ionic headgroups dominate adsorption on charged surfaces—a finding validated through independent batch experiments.

This work shifts the paradigm from material-specific experimentation to physics-guided sorbent design, offering a computationally efficient tool to accelerate the development of multifunctional materials for PFAS remediation. By bridging mechanistic depth with data-driven agility, the framework empowers researchers to rationally engineer sorbents that leverage synergistic adsorption-absorption mechanisms, advancing sustainable water treatment technologies in the face of evolving regulatory and environmental demands.

References:

[1] Kazi M-K, Varghese S, Sarker N, Aich N, Gadhamshetty V. Advancing PFAS Remediation through Physics-Based Modeling of 2D Materials: Recent Progress, Challenges, and Opportunities. Industrial & Engineering Chemistry Research. 2025;64:1894-906.

[2] He Y, Cheng X, Gunjal SJ, Zhang C. Advancing PFAS Sorbent Design: Mechanisms, Challenges, and Perspectives. ACS Materials Au. 2024;4:108-14.

[3] Park M, Wu S, Lopez IJ, Chang JY, Karanfil T, Snyder SA. Adsorption of perfluoroalkyl substances (PFAS) in groundwater by granular activated carbons: Roles of hydrophobicity of PFAS and carbon characteristics. Water Research. 2020;170:115364.

[4] Stebel EK, Pike KA, Nguyen H, Hartmann HA, Klonowski MJ, Lawrence MG, et al. Absorption of short-chain to long-chain perfluoroalkyl substances using swellable organically modified silica. Environmental Science: Water Research & Technology. 2019;5:1854-66.