Staged computational prioritization of lanmodulin binding-site designs for dysprosium-focused rare-earth separation

Rare-earth separations are energy- and solvent-intensive, motivating aqueous protein-based alternatives. We developed a staged computational workflow combining four experimental lanmodulin templates, constrained ProteinMPNN and LigandMPNN sampling, score-only Rosetta triage, and an exploratory OpenMM geometry screen to prioritize a candidate for future reduced-cluster multireference characterization. The residue-role inventory comprised 70 category memberships representing 66 unique positions because four positions belonged to two roles. Fixed-seed sequence generation and subsequent triage yielded three finalist models. Three 100 ps OpenMM replicas were evaluated for each candidate-metal combination, yielding 45 simulations. Audit of the retained systems showed that amber19-all.xml and amber19/opc3.xml were loaded, but no explicit bulk-solvent molecules or counterions were present. The retained system XMLs and execution code contained a standard NonbondedForce and no verified C4/r4 implementation; the calculations therefore cannot be described as verified 12-6-4 simulations. Every replica produced a 3.2 Å any-oxygen occupancy fraction of 1.000 using oxygen atoms present in the retained topology, making this descriptor non-discriminating. The Hans pocket and AM1/Mex descriptors each averaged four target-metal sites, whereas the Hans interface descriptor averaged 14 sites, including 10 on fixed chains B-D. Cross-topology distance comparisons were therefore not site matched. Within each topology, Al(III) and Fe(III) produced shorter model-dependent distances than Dy(III), Nd(III), and Y(III), but these observations are not binding free energies or evidence of selectivity. A dimensionless Composite Quality Index, CQI = 0.7L + 0.3G, ranked the Hans chain-A pocket model highest at 0.802 and was used solely for candidate prioritization. The workflow therefore identifies a model-prioritized candidate for further computation and experiment.