Machine learning dihydrogen activation in the chemical space surrounding Vaska's complex

Area of Science:

• Catalysis
• Computational Chemistry
• Materials Science

Background:

• Homogeneous catalysis with transition metal complexes is vital for organic synthesis, water splitting, and CO2 reduction.
• Ligand design is crucial for tuning catalytic activity but faces challenges due to vast possibilities and complex interactions.
• Vaska's complex serves as a model system for studying hydrogen activation via oxidative addition.

Purpose of the Study:

• To develop a computational framework combining DFT and ML for predicting reactivity in large chemical spaces.
• To accelerate the discovery of optimal ligands for homogeneous catalysis.
• To identify key features governing hydrogen activation barriers in Vaska's complex derivatives.

Main Methods:

• Density Functional Theory (DFT) calculations were used to generate data for training ML models.
• Machine learning models, including Bayesian-optimized Artificial Neural Networks (ANNs) and Gaussian Process (GP) models, were employed.
• Feature engineering using autocorrelation, deltametric functions, and fingerprints was utilized.
• Gradient Boosting (GB) was used for feature selection and model interpretation.

Main Results:

• ML models accurately predicted hydrogen (H2)-activation barriers for 2574 Vaska's complex derivatives.
• ANN models achieved mean absolute errors (MAE) of 1-2 kcal mol-1.
• A GP model further enhanced accuracy, achieving MAE below 1 kcal mol-1 with minimal data.
• Key features influencing H2-activation included chemical composition, atom size, and electronegativity.

Conclusions:

• Combining DFT and ML offers a rapid and accurate approach to predict reactivity in homogeneous catalysis.
• This methodology significantly accelerates the exploration of large chemical spaces for catalyst design.
• The study identified critical ligand features for optimizing hydrogen activation, paving the way for novel catalyst development.