Nonparametric Bayesian inference for meta-stable conformational dynamics

Area of Science:

• Biophysics
• Computational Biology
• Statistical Mechanics

Background:

• Analyzing biomolecular structural dynamics is crucial for understanding complex molecular systems.
• Both experimental and computational methods generate large datasets with an unknown number of underlying molecular states.
• Conformational switching is a key phenomenon studied in biomolecular dynamics.

Purpose of the Study:

• To develop a novel generative, Bayesian nonparametric hidden Markov state model for analyzing biomolecular dynamics.
• To address challenges of large data volumes and unknown numbers of distinct molecular states.
• To provide a scalable and computationally tractable method for analyzing conformational switching.

Main Methods:

• Utilized hierarchical Dirichlet processes to define distinct Markov states without prior specification of their number.
• Employed a mean-field variational inference approach for scalable inference on large datasets.
• Developed a computationally tractable approximation for analyzing angular data, a common data type in this field.

Main Results:

• The proposed model successfully handles large datasets and infers the number of hidden states.
• Demonstrated the method's utility on synthetic data, benchmark problems, and real-world electrophysiological data from an ion channel.
• The model effectively captures the conformational switching dynamics of biomolecules.

Conclusions:

• The developed Bayesian nonparametric hidden Markov state model offers a powerful and scalable approach for analyzing biomolecular structural dynamics.
• This method advances the study of conformational switching by providing robust inference from large and complex datasets.
• The model has practical utility in both computational and experimental biophysics research.