A Probabilistic Framework for Constructing Temporal Relations in Replica Exchange Molecular Trajectories.

Area of Science:

• Biophysics
• Computational Biology
• Machine Learning

Background:

• Understanding biomolecular structure and dynamics is crucial for biological mechanisms.
• Stochastic biological processes yield unique simulation trajectories, complicating direct analysis.
• Existing statistical models struggle with trajectories lacking temporal data, such as from Monte Carlo or replica exchange molecular dynamics (REMD).

Purpose of the Study:

• To develop a novel probabilistic algorithm for extracting reactive pathways from molecular dynamics trajectories, even without temporal information.
• To enable analysis of simulations from methods like REMD and Monte Carlo.
• To provide a tractable method for understanding biomolecular mechanisms.

Main Methods:

• Representing trajectory frames as vectors of interaction and conformational energies.
• Applying Principal Component Analysis (PCA) for dimensionality reduction.
• Clustering frames into metastable states using a density-based algorithm.
• Constructing a graph of metastable states with edges learned via expectation-maximization.
• Identifying the most reactive pathway as the widest path in the learned graph.

Main Results:

• The algorithm successfully extracted reactive pathways from molecular trajectories lacking temporal data.
• The method was validated on an RNA hairpin unfolding trajectory in urea solution.
• The approach facilitated a more tractable understanding of the RNA hairpin unfolding mechanism.

Conclusions:

• The proposed probabilistic algorithm effectively identifies biomolecular reactive pathways from time-independent trajectory data.
• This method expands the applicability of trajectory analysis to simulations that do not inherently provide temporal information.
• The approach offers a powerful tool for elucidating complex biomolecular mechanisms, such as protein and RNA unfolding.