A Robotics Approach to Enhance Conformational Sampling of Proteins

Proteins are biological macromolecules that play essential roles in living organisms. Furthermore, the study of proteins and their function is of interest in other fields in addition to biology, such as pharmacology and biotechnology. Understanding the relationship between protein structure, dynamics and function is indispensable for advances in all these areas. This requires a combination of experimental and computational methods, whose development is the object of very active interdisciplinary research. In such a context, this paper presents a technique to enhance conformational sampling of proteins carried out with computational methods such as molecular dynamics simulations or Monte Carlo methods. Our approach is based on a mechanistic representation of proteins that enables the application of efficient methods originating from robotics. The paper explains the generalities of the approach, and gives details on its application to devise Monte Carlo move classes. Results show the good performance of the method for sampling the conformational space of different types of proteins. INTRODUCTION Proteins are essential components of living organisms. They have a wide range of functions into cells such as catalysis, regulation, signaling, transport, storage and structural functions. In addition to their primary importance in biology, proteins are also key items in other domains. Indeed, proteins are pharmaceutical targets and drugs, their catalytic properties are exploited in biotechnology, and they are used as components of nano-devices in the rising field of bionanotechnology. The study of the relationship between structural and dynamic features of proteins and their function is fundamental in all these domains. Unfortunately, Address all correspondence to this author. experimental methods to provide accurate, atomic-scale data for such studies are limited and expensive. Therefore, computational methods are being developed since several decades ago to model proteins and to simulate their behavior. Protein modeling is very challenging because of the large size and the flexibility of these biological molecules. Indeed, an appropriate model of a protein should not involve a single structure, but a set of conformational ensembles. Given a conformational state of a protein, obtained by experimental techniques like X-ray crystallography [1] or by structure prediction algorithms [2], several methods can be applied to explore the conformational space aiming to provide a conformational ensemble suitable for the analysis of physico-chemical properties. Most of these methods are based on molecular dynamics simulations or on the Monte Carlo method [3, 4]. Nevertheless, alternative methods have been proposed in recent years, some of which are originating from robotics [5]. This paper presents an approach to enhance conformational exploration methods. The approach is based on a mechanistic view of proteins. The idea is to cut the protein into small fragments, called tripeptides, that can be represented as kinematic chains, similar to robotic manipulators. Such a representation enables to efficiently perform local deformations of the protein using semi-analytical inverse kinematics methods. The present work focuses on a specific application of this approach for devising Monte Carlo move classes. Nevertheless, the tripeptidebased protein representation introduced below could be exploited within other types of methods. The Monte Carlo (MC) method [4, 6], explores the conformational space through a random walk. At each iteration, the protein conformation is randomly perturbed, and the trial move is accepted or rejected with a probability that depends on the potential energies of the old and the new states. The main difficulty involving the application of the MC method to proteins consists