a metropolis monte carlo algorithm to compute low
TRANSCRIPT
A Metropolis Monte Carlo Algorithm to compute Lowenergy Structures of an RNA chain
Rachael Chung UNC‐Charlotte
9201 University City Blvd Charlotte, NC 28223 [email protected]
Amarda Shehu, PhD. George Mason University 4400 University Drive Fairfax, VA 22030 [email protected]
Beenish Jamil George Mason University 4400 University Drive Fairfax, VA 22030 [email protected]
ABSTRACT This paper will be discussing a program that calculates low energy, tertiary RNA structures. We make use of a Metropolis Monte Carlo search algorithm to find these low energy structures. We use the Metropolis criterion [1] to cross through an energy landscape to get as close to the overall minimum potential energy as possible. The importance of finding the structure of RNA is understandable as the structure implies function. Much of RNA function is still unknown and very mysterious. Structures created by the program will be compared to existing natural structures to test accuracy and function. Keywords Metropolis, Monte Carlo, RNA, tertiary, NAST, PDB, dihedral rotations. INTRODUCTION Ribonucleic acid (RNA) is a biological molecule essential to life and the basic biology of organisms. It can be found in all parts of a life form. RNA molecules assume various structures and functions in cells. Functions can range from protein synthesis to carrying genetic information, like in viruses. However, not all RNA functions are understood or discovered. In most computational biology, RNA research is on determination and understanding of secondary structure, this work focuses on computing tertiary RNA structures. We address the determination of RNA tertiary structure from knowledge of its nucleotide, or the building block, sequence. This is important as; the structure can determine biological function. RESULTS The program is successfully able to predict tertiary RNA structures with low – energy confirmations. We are able to use given information about a particular RNA nucleotide sequence and its secondary structure, to make a preliminary tertiary
structure. That structure is then modified and its dihedrals are rotated to create a structure with a lower energy than what it started with. Ultimately, a lower potential energy for the structure is reached after runtime. We were
successfully able to implement a Monte Carlo search algorithm. Within the algorithm, we were able to execute a Metropolis criterion [1] that helped accept or reject potential structures and navigate through an energy landscape. MATERIALS AND METHODS We used a program called NAST, Nucleic Acid Simulation Tool, [2] to implement a Metropolis Monte Carlo search for low‐energy tertiary structures. NAST [2] provides the capability to model an RNA molecule in coarse‐grained detail as a long chain of one pseudo atom per nucleic acid, meaning the pseudo bonds are preserved and act as the center of rotation in dihedral changes. We read in information on a particular RNA strand from a PDB (Protein Data Bank) file. The PDB is a main source for obtaining 3‐D data on proteins and nucleic acids. NAST [2] then uses the known secondary structures from the PDB file to put together and predict tertiary structures. It is also able to measure the RNA’s potential energy. In order to calculate and ensure that we have the most energy efficient structure, we implemented an algorithm for a Metropolis Monte Carlo search. Similar search algorithms for protein tertiary structures inspired our Metropolis Monte Carlo
Figure 1:Tertiary RNA interferase. Example of tertiary folding. (copyright scienceart.com)
algorithm. A Monte Carlo algorithm repetitively compares arbitrary samples in order to obtain the result. It is a local search method meaning, it moves from one solution to another until either it finds the best solution or the amount of given time runs out. In our case, it takes different strands of the same RNA and compares them to each other to find the most energy efficient structure. To obtain the different strands of RNA, we alter the angles between different molecules in the RNA. In essence, the algorithm explores the space of dihedral
Figure 2: Dihedral angle Shows three vectors. Each circle represents an RNA
molecule and is number 1‐4 from left to right.
degrees of freedom of an RNA chain. Dihedral degrees or angles combine three vectors from an RNA chain. The vectors in our case are made up of the bonds between four consecutive molecules in the RNA chain. The distances between the first and the third as well as the second and the fourth molecules are found; making two planes. The angle between these two planes is what makes up the dihedral angles in the RNA chain. The algorithm looks for rotations or angle changes that result in low‐energy structures. Our algorithm searches through a chain of RNA and slightly changes the angle of a dihedral through use of a quaternion‐based rotation. After each dihedral rotation we visualized the result via VMD, Visual Molecular
dynamics. By doing this we were able to compare our result to our original structure. We could see more clearly how the structure had changed and see how the distances between the RNA molecules made a difference in the potential energy of the structure. Conformations are accepted or rejected through the Metropolis criterion [1]. The Metropolis criterion [1] employs the NAST [2] coarse‐grained potential energy function. The energy function works with course grained RNA structures and is based on NAST’s energy function [2]. The Metropolis criterion [1] allows for a new structure conformation to be accepted with some probability that reflects a change after the new conformation is made. It ensures that the lowest energy is obtained without getting stuck at a local minimum in an energy landscape. Energy landscapes are not flat; rather they have many steep valleys and peaks creating local minima and maxima, as well as, an overall minima and maxima. The criterion [1] allows us to give and take which high and low energy structure to accept. It allows us to traverse through areas of local minima into areas that increase in energy in an effort to reach the overall minima. As a result, our program is able to obtain low‐energy tertiary structures of an RNA chain. DISCUSSION Ongoing and future work will focus on enhancing the sampling capability of our method so far. We will check the results of our algorithm with tertiary structures found in nature in order to benchmark the functions. Obtaining novel structures will greatly increase our capability to elucidate potentially novel biological functions of RNA molecules. We will also invest time in the exploration of new ways to find tertiary structures. We plan to find new ways in which to manipulate the potential RNA structures so that we can continue to work towards an overall minimum potential energy. We will also expand the basic Metropolis algorithm via simulated annealing to enhance the sampling of RNA. REFERENCES [1] Metropolis, Nicholas; Rosenbluth, Arianna W.; Rosenbluth, Marshall N.; Teller, Augusta H.; Teller, Edward (1953). "Equation of State Calculations by Figure 3: A visual representation of an RNA
tertiary structure before (grey) and after (green) the execution of our algorithm (through use of VMD).
Fast Computing Machines". Journal of Chemical Physics 21 (6): 1087‐1092. [2] Jonikas MA, Radmer RJ, Laederach A, Das R, Pearlman S, Herschlag D, Altman RB. Coarse‐grained modeling of large RNA molecules with knowledge‐based potentials and structural filters. RNA. 2009 Feb;15(2):189‐99. PMID: 19144906 (2009)
ACKNOWLEDGEMENTS I would like to thank CRAW – DREU for giving me the opportunity to work on this research with my mentor Dr. Shehu and fellow undergraduate researcher, Beenish Jamil.