Generating transition states of isomerization reactions with deep learning

Area of Science:

• Computational Chemistry
• Chemical Kinetics
• Machine Learning Applications

Background:

• Quantitative understanding of reaction kinetics is limited by poor quality and generation of kinetic data.
• Conventional methods for generating transition state structures lack speed, accuracy, or scope.

Purpose of the Study:

• To present a novel computational method for generating 3D transition state structures for isomerization reactions.
• To overcome the limitations of existing methods in terms of speed, accuracy, and scope.

Main Methods:

• Utilized a graph neural network to predict the transition state distance matrix from reactant and product geometries.
• Employed least squares optimization to reconstruct 3D coordinates based on the predicted distance matrix.
• Validated the generated transition state structures using a quantum mechanics workflow.

Main Results:

• The novel method successfully generated viable geometries for transition states.
• The algorithm achieved excellent accuracy in predicting transition states.
• The approach demonstrated superior performance compared to conventional methods.

Conclusions:

• The developed method offers a significant advancement in computing chemical reactions.
• Combining neural networks with quantum chemistry calculations presents a promising future for reaction kinetics studies.
• This approach is expected to become a preferred method for computing chemical reactions.