Title: Manifold Diffusion for Structure Generation of Transition Metal Complexes

Abstract:

Transition metal complexes play a pivotal role in materials science, drug discovery, and catalysis, with their functional properties heavily dependent on their three-dimensional structural arrangements. Nevertheless, generating accurate structures for these molecules remains difficult due to their complex electronic characteristics and atypical bonding configurations. To address this, we present TMCgen, a novel machine learning framework based on manifold diffusion that facilitates the rapid and precise generation of transition metal complex geometries. TMCgen optimizes the diffusion process by targeting the critical geometric degrees of freedom inherent to these complexes: it models diffusion across metal-ligand coordination angles while simultaneously applying torsional and rotational diffusion to the ligands. The model demonstrates robust capabilities in producing accurate coordination environments across a wide variety of experimentally validated organometallic and bioinorganic systems. Notably, TMCgen achieves this efficiency with a minimal number of inference steps. These findings highlight the promise of manifold-based generative approaches for data-efficient structural modeling, opening new avenues for the property-driven design of transition metal complexes.

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC