February 2026 – Present
Undergraduate Researcher
- Combined lanthanide(III) binding data from 5 sources, filled in missing ligand structures from PubChem, and built features from Morgan fingerprints, RDKit descriptors, and metal properties.
- Trained a tensor-network kernel machine plus LightGBM model, validated with 10-fold CV grouped by ligand, and a second model to isolate metal selectivity, followed by a SHAP analysis to see which features drive predictions.
- Screened ligands across the 4f series for selectivity, and then used a DEAP genetic algorithm to mutate top ligands into new candidates, scored on affinity, selectivity, and 3D strain.