University of Cambridge > > Machine Learning @ CUED > Probabilistic methods for biomolecular structure simulations

Probabilistic methods for biomolecular structure simulations

Add to your list(s) Download to your calendar using vCal

If you have a question about this talk, please contact Zoubin Ghahramani.

Biomolecules, such as RNAs and proteins, play fundamental roles in all living cells. Understanding the exact function of these molecules is of great importance in biology and medicine, and this understand often requires knowledge of the molecules’ structure in atomic detail. As experimental structure determination can be difficult and laborious, there is a great interest in computational methods for biomolecular structure simulation. In this talk I will present a number of probabilistic methods that address challenges in Markov chain Monte Carlo (MCMC) based structure simulations.

In the first part of the talk, I will present a probabilistic model of RNA conformational space. The current state-of-the-art in conformational sampling is based on fragment assembly. However, the discrete nature of fragments necessitates the use of carefully tuned, unphysical energy functions, and their non-probabilistic nature impairs unbiased sampling. I will present a solution to the sampling problem without these limitations: a probabilistic model of RNA structure that allows for efficient sampling of RNA conformations in continuous space, and with associated probabilities. The method provides both a theoretical and practical solution for a major bottleneck on the way to routine RNA structure prediction in atomic detail.

In the second part of the talk, I will present a generalized ensemble MCMC algorithm. The standard Metropolis-Hastings algorithm can suffer from the generic deficiency of slow convergence and poor mixing. In biomolecular simulations this problem is normally address by using generalized ensembles. I will present an automated histogram based method for estimating generalized ensemble weights. In our method, the density of states is estimated iteratively using a maximum likelihood approach. The method scales to high dimensional systems due to the introduction of an automated binning procedure. The strength of the method is demonstrated, by applying the method to simulate the folding of a small polypeptide.

This talk is part of the Machine Learning @ CUED series.

Tell a friend about this talk:

This talk is included in these lists:

Note that ex-directory lists are not shown.


© 2006-2020, University of Cambridge. Contact Us | Help and Documentation | Privacy and Publicity