(311d) Bayesian Optimization of Lennard-Jones Forcefields for Aqueous Lanthanide Ions

Lanthanides (often referred to as rare earth elements, or REEs) are a common component of many consumer electronics. Due to their similar physiochemical properties they are notoriously difficult to purify from their natural state. This has led to worldwide awareness and the fear that demand could soon outstrip the supply. Current state-of-the-art extraction techniques involve several extraction steps, each utilizing solvents that are hazardous and environmentally unfriendly. In search of a more sustainable extraction method, success has been made in the direction of "bio-mining," which exploits various biomolecular structures and their natural interaction with ionic species. These novel peptides and the interplay between their structure and their REE binding affinity are not yet well-understood. Molecular modeling can help shed light on binding energies and conformational changes that are critical in the binding process, but this requires confidence on the empirical forcefields used to describe each chemical species. Unfortunately, classical interaction parameters for monoatomic trivalent ions (such as the lanthanides) are known to suffer from deficiencies due to their high charge density which leads to local polarization. We demonstrate that Bayesian optimization can be implemented to derive parameters that reproduce a collection of experimental observables to produce improved models of REEs and allow for accurate property prediction, thereby enabling improved design of sustainable separation technologies.