42: JACOBIK: Inverse kinematics for transition metal complex conformer generation

Abstract: Efficient and robust 3D structure generation remains a critical hurdle for modern computational discovery of transition metal complexes (TMCs). Existing approaches range from fast template-based methods to accurate quantum calculations, but neither reliably handles polydentate or mixed-ligand topologies at scale. However, the uniquely constrained conformations TMCs exhibit make them a natural fit for inverse kinematics approaches. We present JACOBIK, a conformer generation tool for TMCs based on Jacobian inverse kinematics. By decomposing ligands into kinematic chains along rotatable pivot bonds and rigid bodies, JACOBIK is able to solve for various conformations of TMCs directly via a set of statistically derived constraints. Geometric constraints are derived from Gaussian Mixture Models trained on the tmQM dataset across 4,846 distinct metal-donor environments and 49,535 donor-pair angle contexts. The models are fit to distributions over six extracted geometric quantities capturing both standard conformational preferences and complex effects such as Jahn-Teller distortions and trans influence, without requiring explicit electronic structure calculations. A parallel tree search identifies the most probable conformers consistent with these constraints, assembling them into an ensemble. Preliminary benchmarks demonstrate speeds 2-3 orders of magnitude faster than existing quantum methods and robustness across a much broader class of ligands than molSimplify. The speed and generalizability demonstrated position JACOBIK as a practical structure generation layer for high-throughput TMC discovery pipelines.