Ligand additivity relationships enable efficient exploration of transition metal chemical space

N. Arunachalam, Stefan Gugler, Michael G. Taylor, Chenru Duan, Aditya Nandy, Jon Paul Janet, Ralf Meyer, Jonas Oldenstaedt, Daniel B. K. Chu, Heather J. Kulik

TL;DR

This paper tackles the challenge of efficiently exploring the vast chemical space of transition metal complexes. By leveraging the concept of "ligand additivity" and developing sophisticated interpolation methods, the authors demonstrate a powerful strategy to predict properties of complex molecules, significantly accelerating the discovery of new materials.

What Problem Does This Paper Address?

The sheer number of possible transition metal complexes is astronomical, making exhaustive computational screening infeasible. Existing methods often struggle to navigate this space efficiently, especially for lower-symmetry complexes which are underrepresented in experimental databases. The challenge lies in developing predictive models that can accurately estimate properties of these complex molecules without requiring computationally expensive calculations for every single one. This paper aims to bridge this gap by using ligand additivity to infer properties of complex, heteroleptic complexes from simpler, homoleptic ones.

Methods

The core of this work lies in developing and applying "ligand additivity" principles for property prediction. This involves:

  • Large-scale Curation: Analyzing a substantial dataset of mononuclear octahedral transition metal complexes from the Cambridge Structural Database (CSD).
  • Symmetry and Ligand Analysis: Developing methods to classify complexes by symmetry and identify unique ligands, revealing the vast theoretical space compared to experimentally characterized ones.
  • Interpolation Schemes:
    • Coarse Interpolation: Using homoleptic complexes (where all ligands are the same) as a baseline to estimate properties of heteroleptic complexes (with different ligands).
  • Improved Interpolation: Developing more sophisticated schemes that incorporate information from specific binary complex symmetries (like CS/TS and FS/MS) to better capture isomer effects and improve accuracy.
  • Multi-stage Strategy: Combining coarse interpolation with refined interpolation on targeted subspaces to efficiently identify promising candidates.
  • Property Prediction: Focusing on predicting adiabatic spin-splitting energy (ΔE<sub>H-L</sub>) and HOMO energy levels.

Datasets

  • Cambridge Structural Database (CSD): The primary dataset used for curating a large collection of mononuclear octahedral transition metal complexes. This dataset was filtered to identify "computation-ready" complexes and a subset of Fe(II) complexes.
  • Newly Synthesized/Calculated Complexes: The paper validates its approach by predicting and then implicitly validating properties for 9 new binary and ternary complexes.

Key Insights and Ideas

  • Vast Unexplored Space: The theoretical chemical space of transition metal complexes is orders of magnitude larger than what has been experimentally characterized, highlighting the need for efficient exploration strategies.
  • Ligand Additivity is Powerful: The properties of complex heteroleptic complexes can be reasonably inferred from the properties of simpler homoleptic complexes, forming the basis for predictive models.
  • Isomers Matter: Simple linear averaging of homoleptic properties is insufficient; incorporating information about cis/trans isomerism and other symmetry-related effects significantly improves prediction accuracy.
  • Data-Efficient Discovery: A multi-stage approach, starting with coarse interpolation and refining with more complex models on targeted subspaces, offers a computationally efficient way to discover novel compounds with desired properties.

Key Results

  • The theoretical space of Fe(II) complexes with a small number of unique ligands is estimated to be around 816,000, vastly exceeding the experimentally characterized ones.
  • Improved interpolation schemes incorporating isomer effects predict adiabatic spin-splitting energy with an accuracy of ~2 kcal/mol and HOMO levels with an accuracy of <0.2 eV.
  • The developed multi-stage strategy successfully identified promising complexes within a targeted property region, demonstrating its utility for efficient chemical space exploration.
  • Errors for ΔE<sub>H-L</sub> were reduced from ~5 kcal/mol with homoleptic averaging to ~2 kcal/mol with FS/MS-interpolation.

Limitations and Future Work

  • Ligand Compatibility: The study acknowledges that not all ligand combinations are sterically or electronically feasible, which is not fully accounted for in theoretical enumeration.
  • Generalizability: While demonstrated for magnetic and orbital energy properties, the authors suggest this approach could be extended to other properties like redox potentials or catalytic activity.
  • Integration with ML: The additivity approach could be integrated with machine learning models to further enhance property prediction and dataset augmentation.
  • Further Refinement: The authors suggest that further refinement of interpolation schemes, potentially incorporating more complex ligand interactions, could lead to even higher fidelity predictions.

Summary updated

Abstract

To accelerate exploration of chemical space, it is necessary to identify the compounds that will provide the most additional information or value. A large-scale analysis of mononuclear octahedral transition metal complexes deposited in an experimental database confirms an under-representation of lower-symmetry complexes. From a set of around 1000 previously studied Fe(II) complexes, we show that the theoretical space of synthetically accessible complexes formed from the relatively small number of unique ligands is significantly (ca. 816k) larger. For the properties of these complexes, we validate the concept of ligand additivity by inferring heteroleptic properties from a stoichiometric combination of homoleptic complexes. An improved interpolation scheme that incorporates information about cis and trans isomer effects predicts the adiabatic spin-splitting energy to around 2 kcal/mol and the HOMO level to less than 0.2 eV. We demonstrate a multi-stage strategy to discover leads from the 816k Fe(II) complexes within a targeted property region. We carry out a coarse interpolation from homoleptic complexes that we refine over a subspace of ligands based on the likelihood of generating complexes with targeted properties. We validate our approach on 9 new binary and ternary complexes predicted to be in a targeted zone of discovery, suggesting opportunities for efficient transition metal complex discovery.
BibTeX
@article{arunachalam2022ligand,
  author = {N. Arunachalam and Stefan Gugler and Michael G. Taylor and Chenru Duan and Aditya Nandy and Jon Paul Janet and Ralf Meyer and Jonas Oldenstaedt and Daniel B. K. Chu and Heather J. Kulik},
  title = {{Ligand additivity relationships enable efficient exploration of transition metal chemical space}},
  journal = {The Journal of Chemical Physics},
  year = {2022},
  doi = {10.1063/5.0125700},
}
Figure