Identifying Dynamic Metal-Ligand Coordination Modes with Ensemble Learning.

Area of Science:

• Organometallic Chemistry
• Computational Chemistry
• Materials Science

Background:

• Understanding ligand coordination to metals is vital for transition metal complexes (TMCs).
• Existing analyses often overlook dynamic, hemilabile coordination modes common in catalysis.
• Hemilability, where a ligand can bind and unbind dynamically, presents a challenge to traditional coordination assumptions.

Purpose of the Study:

• To analyze trends in ligand coordination modes, specifically hemilability.
• To develop a computational method for classifying hemilabile ligands and predicting coordination modes.
• To accelerate the discovery of novel transition metal complexes.

Main Methods:

• Curated datasets of hemilabile and nonhemilabile ligands from the Cambridge Structural Database.
• Trained graph neural networks for accurate classification of hemilabile ligands.
• Developed an ensemble algorithm to predict primary and alternative coordination modes from SMILES strings.
• Utilized density functional theory (DFT) to calculate energy differences for novel TMCs.

Main Results:

• Introduced four exhaustive and mutually exclusive types of hemilability.
• Achieved high accuracy, precision, and recall in classifying hemilabile ligands using graph neural networks.
• Successfully predicted primary and alternative coordination modes for novel TMCs.
• DFT calculations validated the plausibility of predicted alternative coordination modes.

Conclusions:

• The developed algorithm accurately predicts ligand coordination modes, including hemilability.
• This work provides open-source workflows to accelerate the discovery of novel TMCs.
• The findings will aid both experimental and virtual screening campaigns in organometallic chemistry.