Following the Committor Flow: A Data-Driven Discovery of Transition Pathways

Area of Science:

• Computational Chemistry
• Molecular Dynamics
• Reaction Mechanism Discovery

Background:

• Identifying rare events and transition pathways in molecular systems remains a significant challenge.
• The committor probability is a key metric for defining reaction coordinates in molecular simulations.
• Ensuring consistency between transition pathways and the committor function is crucial for mechanistic accuracy.

Purpose of the Study:

• To develop an iterative framework for inferring the committor probability.
• To identify and refine the most relevant transition pathways in molecular systems.
• To enable accurate estimation of reaction rate constants.

Main Methods:

• An iterative approach using biased sampling and neural network approximation of the committor probability.
• Extraction of dominant transition channels from learned committor isocommittor surfaces.
• Iterative refinement of committor and transition paths until convergence.

Main Results:

• The proposed framework successfully infers the committor probability and identifies key transition pathways.
• Demonstrated effectiveness on benchmark systems: 2D model potential, peptide transitions, Diels-Alder reaction, and Trp-cage folding.
• The refined committor enables accurate reaction rate constant estimation.

Conclusions:

• The iterative framework provides a robust method for discovering transition pathways and understanding reaction mechanisms.
• This approach enhances molecular dynamics simulations by improving the accuracy of committor estimation.
• The method offers a significant advancement in the study of rare events in molecular systems.