Continued Sampling: Using Data Aware Feedback and Control for Peptide Conformational Search

Abstract

Sampling till completion is a new framework for conformational search that uses a coupled feedback between running computational dynamics and a large database. By using this feedback we can get improved convergence as well as more efficient sampling. We have applied this concept to sampling of peptide degrees of freedom and the first part of our presentation will highlight the challenges in making this idea work by addressing both hardware and software issues. This includes our efforts to create rapid query evaluations and our use of coupled CPU and GPU systems. The second part of our presentation will focus on the nature of the efficiency gains and how the concept may be helpful for sampling within the Onsager-Machlup framework for understanding conformational transitions. This enables us to view certain degrees of freedom as more important for the transitions than others and ties into dynamic importance sampling and effective transfer entropy approaches.