CAREER: Reinforcement Learning of the Free Energy Landscapes of Proteins

点击翻译按钮获取中文摘要

Proteins are molecular machines that perform a variety of tasks in living organisms, such as transporting nutrients and communicating signals within and between cells. Some of these proteins undergo a change in their conformation or shape to carry out these functions. It is challenging to visualize the different shapes of these molecular machines using traditional experimental techniques, such as X-ray crystallography. Therefore, the molecular level understanding of these processes and its implications for protein function remains elusive. This lack of molecular information makes it difficult to develop engineering approaches to regulate protein function. Molecular Dynamics simulations provide a way to observe these movements at the atomic scale. However, significant amount of computer time is required to observe long timescale conformational change processes in proteins. The objectives of this project are (1) to develop efficient algorithms based on machine learning techniques to understand how proteins undergo conformational change and (2) to apply these algorithms to understand the role of protein dynamics in transport of molecules across the cell membrane via membrane transporters. The specific transporters investigated in this project play a critical role in determining crop productivity and neurological disorders in humans. In concert with these research objectives, the PI will develop outreach activities teaching high school girls about computational methods used to investigate protein function via Girl's Adventure in Mathematics, Engineering and Sciences summer camp at the University of Illinois. PI also plans to engage African-American boys at local Urbana-Champaign schools via a three-day after-school program to teach them about protein structure and function.This project will develop computational methods that can efficiently explore the free energy landscapes associated with protein conformational changes. This work is guided by the hypothesis that leveraging ideas from reinforcement learning technique and using evolutionary coupled residue pair distances as order parameters for protein functional dynamics will allow efficient sampling of free energy landscapes. The development of the algorithm called "Reinforcement Learning-Based Adaptive Sampling" (REAP) has been initiated. Preliminary results have shown promising application of these ideas to several proteins. The fully developed algorithm would be particularly useful for systems with limited structural information, as order parameters can be identified using evolutionary coupled residues based on sequence information alone. The specific goals of this project include: (1) further develop the REAP methodology to efficiently explore the free energy landscapes associated with protein function, (2) test the new methodology to understand molecular processes of high biological importance but with limited availability of structural information. In particular, molecular mechanisms of substrate transport and its regulation for the following transport processes will be investigated: (1) Nitrate transport process via root-associated transporters in plants with applications in increasing the crop yields. (2) Serotonin transport in brain via human Serotonin transporter for elucidating molecular origin of neurological disorders. (3) Sugar transport via SWEET family transporter in Rice with applications in enhancing plant growth. Despite differences in their function and structures, these systems share similarities in terms of their modes of regulation that could allow for a comprehensive understanding of the regulatory mechanisms in membrane transporters.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.