Exploring the Global Reaction Coordinate for Retinal Photoisomerization: A Graph Theory-Based Machine Learning...

Area of Science:

• Computational chemistry
• Photochemistry
• Machine learning

Background:

• Photoinduced reactions are complex and challenging to study.
• Nonadiabatic molecular dynamics (NAMD) combined with graph theory-based machine learning aids in determining reaction pathways.
• NAMD simulations are computationally intensive due to the need for frequent nonadiabatic coupling vector calculations.

Purpose of the Study:

• To investigate ab initio molecular dynamics (AIMD) as a computationally efficient alternative to NAMD for studying photoinduced reaction pathways.
• To determine a plausible global reaction coordinate for retinal photoisomerization using AIMD.
• To compare the efficiency and results of AIMD and NAMD for studying reaction coordinates.

Main Methods:

• Utilizing ab initio molecular dynamics (AIMD) with appropriate initial conditions.
• Applying graph theory-based machine learning.
• Analyzing internal coordinates and their mutual information (MI) with HOMO energy.
• Comparing AIMD and NAMD simulation results.

Main Results:

• AIMD simulations can effectively determine global reaction coordinates for photoinduced reactions.
• AIMD and NAMD yield similar trends when ranking internal coordinates based on mutual information with HOMO energy.
• The AIMD-based machine learning protocol is 1.5 times faster than NAMD for studying reaction coordinates.

Conclusions:

• AIMD provides a computationally viable and efficient alternative to NAMD for unraveling reaction pathways of photoinduced reactions.
• This methodology is applicable to crucial biological processes like retinal photoisomerization.
• The study highlights a significant speed-up in computational cost for studying reaction coordinates.