Accurate transition state generation with an object-aware equivariant elementary reaction diffusion model.

Area of Science:

• Computational Chemistry
• Artificial Intelligence
• Chemical Reaction Dynamics

Background:

• Accurate transition state (TS) structure identification is crucial for understanding chemical reaction mechanisms and networks.
• Traditional methods rely on computationally expensive quantum chemistry calculations, limiting large-scale exploration.

Purpose of the Study:

• To develop a novel, efficient method for generating 3D transition state structures.
• To accelerate the elucidation of reaction mechanisms and the construction of reaction networks.

Main Methods:

• Developed an object-aware SE(3) equivariant diffusion model.
• The model generates reactant, transition state, and product structures simultaneously, respecting physical symmetries and constraints.
• Incorporated a confidence scoring model for uncertainty quantification.

Main Results:

• The diffusion model generates transition state structures in seconds, a significant speedup from hours required by traditional methods.
• Achieved a median root mean square deviation of 0.08 Å compared to true transition states.
• Enabled accurate reaction barrier estimation by selectively applying quantum chemistry calculations to only the most challenging 14% of reactions.

Conclusions:

• The developed AI model offers a highly efficient approach for predicting transition state structures.
• This method has the potential to revolutionize the construction of large-scale reaction networks, especially those with unknown mechanisms.
• Facilitates faster and more accurate chemical mechanism elucidation.