Automated Prediction of Ground State Spin for Transition Metal Complexes and Benchmarking Physics-based Representations

Yuri Cho, Ruben Laplaz, Sergi Velad, Yannick Calvino Alonso, Ksenia Briling, Clémence Corminboeuf

Exploiting crystallographic data repositories for large-scale quantum chemical computations requires rapid and accurate retrieval of all the necessary information (molecular structure, charge, and spin state) directly from crystal structures. Here, we develop a general approach to predict the ground state spin of transition metal complexes, complementing our prior work on determining metal oxidation states and bond order within the cell2mol software. Starting from a previously curated database extracted with cell2mol [1], we construct the TM-GSspin dataset, which contains 2,032 mononuclear first row transition metal complexes and their computed ground state spins. TM-GSspin dataset is analyzed to identify correlations between the structural and electronic features of the complexes and their ground state spins. Leveraging this knowledge, we build a rule-based empirical model as well as interpretative statistical models that assign the ground state spin of transition metal complexes, with a high accuracy of 97%. Our approach provides a practical way to determine the ground state spin of transition metal complexes directly from the crystal structure without additional computations, thus enabling the automated use of crystallographic data for large-scale computations. Moreover, we utilize our diverse dataset to benchmark the performance of various physics-based representations [2] in predicting the molecular properties of transition metal complexes.

Figure 1. General workflow of predicting ground state spin of transition metal complexes.

1. Vela, S., Laplaza, R., Cho, Y., Corminboeuf, C., npj Comput. Mater. 2022, 8, 188.
2. Musil, F., Grisafi, A., Bartók, A. P., Ortner, C., Csányi, G., Ceriotti, M. Chem. Rev. 2021, 121, 9759.

 Yuri Cho