Data Analysis of Villin Headpiece Subdomain Folding Simulations. Jagannath Krishnan Department of Computer Science, Stanford University March 2002 Abstract The Folding@Home project simulates protein folding in atomistic detail using a distributed computing approach. My project explores the folding of the villin headpiece by analyzing data generated by the Pande group at Stanford, using the Folding@Home approach. A total of 226.752 microseconds of folding time were simulated and 21 independent folding trajectories were observed. Starting from an extended state a relaxation to an unfolded state is observed which is accompanied by the occurrence of at least one alpha helix. Plateaus in the energy landscape are also observed in nearly all the folded trajectories till the final energy barrier is crossed by which time the hydrophobic core is formed. Helices are observed to form, break and reform, some being more stable than others. Some of the folded structures were seen to have at least 2 of the 3 helices, and rmsd < 3.0 when compared with the experimentally obtained structure (1VII.pdb). The computationally 'impossible' task of simulating the folding of proteins in atomistic detail is being made possible by the use of a radically new distributed computing approach in the folding@home project. Tens of thousands of computers collaborate to generate megabytes of data as the story of each atom is recorded during the folding process. This project seeks to understand the process of protein folding by analyzing the vast amount of data generated while simulating the folding of the villin headpiece. Introduction Protein folding has been called one of the greatest scientific challenges of our time [1]. The limitation of present day experimental methods in determining the structures of proteins has fueled a great deal of interest in trying to predict the structure of a given protein using computational methods. Ab initio structure prediction is the Holy Grail of this field because it empowers the biologist with the ability to predict the structure of even completely new proteins unlike homology or threading based approaches. Protein folding studies the folding trajectory of protein molecules from expanded to native conformation. Understanding the process of folding will not only enable us to predict structure but will also permit us to synthesize new molecules that fold into desired shapes and has implications in drug design. Until recently ab initio structure prediction has been dismissed as a pipe-dream. This is partly because of the difficulty in modeling the folding process in software but more because the computational prowess to carry out the simulations simply does not exist yet. A recent simulation of the villin headpiece that was reported simulated the events in 1 microsecond of folding time[2]. This project used a massively parallel supercomputer to speed up the process of simulation. This approach has its limitations as the amount of communication between the processors is excessive and limits the speedup that can be achieved. The time simulated using this approach falls short of the folding time of even very small molecules.The Folding@Home project attacks the problem using a distributed computing approach. It is based on the insight that given a very large number of molecules that are simulated, each with slightly different forces from the solvent, the probability that a few of the molecules will overcome the energy barrier to reach the native conformation is very high. Small protein molecules are thought to have no intermediate stages and cross the energy barrier just once. For larger molecules with multiple energy barriers the fact that one of the trajectories has crossed a barrier is used to restart all the trajectories from that state. Thus individual simulations don’t need to communicate with each other and given M processors there is a factor of M speedup in the simulation [3]. The simulations are distributed among tens of thousands of processors worldwide and communicate with a central server that gathers data. The aim of my project is to analyze the simulation data obtained for the villin headpiece and to learn about its folding pathways. Most of the data is present is a MySQL database and has been analyzed using Perl scripts I wrote that connect to the database and generate data which is then examined using Mathematica. The data describes various properties of the molecule at each nano second as the molecule goes from an unfolded to a folded state. In the next subsection the villin headpiece is introduced. Section 2 presents the overall characteristics of the folding process. Section 3 singles out a representative folded trajectory and analyzes it in detail. Section 4 analyzes presents an ensemble level analysis of the trajectories. Section 5 summarizes the results. References are listed in section 6 and acknowledgements follow in section 7. Villin Headpiece Subdomain The Folding@Home project has simulated the folding of the villin headpiece subdomain, which is a 36-residue fast folding protein. It is the C terminal domain of the villin actin binding protein. Being a fast folding protein it has been studied both experimentally and by simulation although previously reported simulations are not as long as the one reported in this paper. The molecule being studied has 3 alpha helices joined by short turns (pdbcode 1vII). The experimentally determined structure is shown in figure 1. This figure also shows the four Phe residues, three of which form the hydrophobic core of this molecule. Figure 1. Structure of the villin headpiece subdomain. Note the alpha helices at each terminal and also the single turn helix in the middle. Three of the four Phenylalanine’s come together to form the core.2.Characteristics of the folding process This section elaborates upon the method used to simulate the folding and introduces terminology that is used in describing the results. It also describes the overall manner in which the protein folding data can be viewed. As mentioned in section 1, the basic idea behind Folding@Home is to start numerous simulations in tandem. The forces on the molecule are modeled in atomistic detail and are recalculated on the order of femtoseconds (e-12 seconds). The coordinates of all atoms and properties of the molecule like the radius of gyration, the solvent accessible surface area(SASA), the rmsd from 1VII.pdb, energy etc. are all
View Full Document
Unlocking...