(628a) Data-Driven Design of Polymeric Adsorbent Materials for Ion-Selective Separations

Ion-selective ligand-functionalized polymers adsorbents are an emerging class of materials under development for the separation and recovery of critical minerals from alternative feedstocks, such as lithium and copper ions from aqueous battery and electronics waste. Ligand-functionalized polymer adsorbents affix an ion-binding ligand molecule to an inert polymer backbone. The choice of ligand depends on the composition of the aqueous feedstock, including the target critical mineral and contaminant profile. Edisonian approaches to selective adsorbent development cannot keep pace with the increasing demand for critical minerals required for electrification. Synthesizing and testing novel adsorbent materials on a trial-and-error basis is prohibitively costly and time-intensive.

Precise, rational design of selective adsorbents requires exhaustive knowledge of the interaction parameters between a critical or contaminant mineral ion and organic ligand. Available datasets that tabulate the reactivity between a mineral ion and ligand molecule (ion-ligand) are incomplete and therefore have limited practical use in material design objectives. Beyond the issue of data sparsity, direct translation of the reported reactivity trend from the ideal/simplified ion-ligand system to the real polymeric environment within the adsorbent and real low-grade aqueous stream is challenging.

Here, we demonstrate a data-driven design framework to reduce the number of required experiments and rapidly accelerate the development of ion-selective adsorbents. First, we overcome data sparsity by applying supervised learning models to accurately interpolate missing ion-ligand interaction strength data (ΔG_Ion-Ligand) for 1091 unique ligands and 23 ions (25093 total ion-ligand pairs; 20% of pairs from literature used in training, 80% interpolated with model) and develop a tool to extrapolate interaction strength trends to novel, unobserved ligand chemistries. Second, we find that Linear Free Energy Relationships (LFERs) can be applied to translate the reported/predicted ion-ligand strength trend to the experimentally observed ion-adsorbent reactivity trend (coefficient of determination above 0.80 between LFER predicted and observed ion-adsorbent free energy of adsorption, ΔG_{Ion-Adsorbent}) for 2 acrylate-based adsorbent chemistries. Finally, we demonstrate how the combination of the interpolated ion-ligand dataset and the experimentally determined LFER can be used to predict adsorption selectivity for 8 critical mineral ions (28 ion-ion pairs) after performing only 3 single-ion adsorption experiments for a novel ligand-functionalized adsorbent chemistry. We highlight future opportunities for data-driven approaches to learn relationships between predicted material properties and LFER parameters to further reduce experiment time in adsorbent research and development.

In summary, by combining data interpolation via supervised learning models with empirically determined reactivity translation relationships, data-driven design methods can rapidly accelerate the design of ion-selective polymer adsorbents key to secure critical mineral supply chains.