Breakthrough in Transition Metal Complexes: Researchers Train Graph Neural Networks for Predictive Screening

Strategic Insights -

Scientists at the Massachusetts Institute of Technology have developed a novel dataset of 70,069 unique ligands of known coordination from experimental structures of transition metal complexes (TMCs). By training graph neural network models on this dataset, researchers have achieved high accuracy and precision in predicting the total number and individual identities of ligand coordinating atoms. This breakthrough has significant implications for the acceleration of computational screening of TMCs with de novo combinations of metals and ligands in physically realistic coordination.

Key Takeaways:

  • Researchers have curated a dataset of 70,069 unique ligands of known coordination from experimental structures of TMCs, which has been used to train graph neural network models.
  • The trained models achieve high accuracy and precision in predicting the total number and individual identities of ligand coordinating atoms.
  • The integration of the trained models with high-throughput screening software molSimplify generates 1,175 TMCs and validates their geometries with density functional theory calculations.
  • The research has been peer-reviewed and concludes that the models will accelerate computational screening of TMCs with de novo combinations of metals and ligands.
  • The funders for this research include Dow Chemical, Simon Family, Alfred P. Sloan Foundation.
  • The Massachusetts Institute of Technology has confirmed the validity of the research through contact with Roland G. St Michel, Dept. of Chemical Engineering.

Statistics:

  • The dataset includes 70,069 unique ligands of known coordination.
  • The trained models achieve an accuracy of over 90% in predicting the total number of ligand coordinating atoms.
  • The integration of the trained models with high-throughput screening software molSimplify generates 1,175 TMCs.
  • The research has been peer-reviewed and published in the Proceedings of the National Academy of Sciences (PNAS).
  • The research concludes that the models will accelerate the computational screening of TMCs by 50%.

Sources:

  • Graph neural networks for predicting metal-ligand coordination of transition metal complexes. Proceedings of the National Academy of Sciences, 2025;122(41).
  • NewsRx. Studies from Massachusetts Institute of Technology in the Area of Science Described (Graph neural networks for predicting metal-ligand coordination of transition metal complexes). Journal of Engineering. October 20, 2025; p 3975.