Biomolecular simulations generate many descriptors such as dihedral angles and pairwise distances. Determining the degrees of freedom crucial for structural transitions necessary to study the kinetics of biomolecules is a daunting task. To determine the reaction coordinates for molecular transitions we used machine learning methods. By representing the commitor function using regression methods with cost function being cross-entropy, we tested the efficiency of various machine learning approaches in identifying the kinetically important degrees of freedom using short molecular simulations. We identify the machine learning methods that best predicts the commitor values. Namely, Neural Networks, Kernel methods and cross-entropy method give rise to the best agreement. We tested our approaches against various molecular systems including oligopeptides and oligonucleotides. The important degrees of freedom for conformational transitions are captured successfully for the test systems. Our approach shows great promise to be used as a reaction coordinate for transition path sampling and enhanced sampling methods. It can be used to identify the mechanism of complex conformational transitions that involve non-trivial degrees of freedom.

Automatic determination of collective variables for biomolecular kinetics

ACS Spring Meeting 2021

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.