Rare earth elements, or lanthanides, are critical materials with many applications in clean energy industries such as batteries, semiconductors, and electric vehicles due to their unique luminescent, magnetic, and catalytic properties. Thus there is a growing demand for developing green industrial extraction of these lanthanides. However, efficient separation of heavy lanthanides (e.g., from Europium to Lutetium) with similar chemical properties is challenging and remains an active research area. We hypothesize that molecular level computations, machine learning tools (ML), and the machinery of statistical mechanics can help guide experimental design through ML-enabled molecular modeling. In this study, we have performed molecular dynamics simulations to investigate the selective binding mechanism of the lanthanide binding tag (LBT) peptides and trivalent lanthanide cations in an aqueous solution. The peptide-lanthanide binding free energy landscapes are constructed by running a well-tempered Metadynamics simulation, revealing the hidden complexity of the binding complex conformations. Structural descriptors, including the distance between the lanthanide cation and the coordination residues of peptide and water, are measured from simulation. A neural network model is then trained to predict the experimentally-measured LBT binding free energy with reasonable accuracy. A ML-guided genetic algorithm is employed to systematically search for peptide mutants with improved lanthanide binding selectivity.
This work demonstrates a quantitative and interpretable mapping between the molecular interactions and the selectivity of the lanthanide binding peptides, paving the road for designing novel biomolecules for efficient extractions of rare earth elements.