Amber Workshop - tutorial 8

(Note: These tutorials are meant to provide illustrative examples of how to use the AMBER software suite to carry out simulations that can be run on a simple workstation in a reasonable period of time. They do not necessarily provide the optimal choice of parameters or methods for the particular application area.)
Copyright Ross Walker 2005

Case Study: All Atom Structure Prediction and Folding Simulations
of a Stable Protein (Folding Trp-Cage Peptide)

	>
Initial Linear Structure		NMR Structure (1L2Y) of TRP Cage Peptide

This tutorial is designed as a case study that will show you how to reproduce the work discussed in the following paper:

It is recommended that you read this paper before starting this tutorial. For references to the amino acid sequences and other relevant information you should refer to those papers cited in the above article.

This tutorial consists of this introduction page and seven subsections as follows:

    1) section1.htm - Building the Structure.
    2) section2.htm - Creating the prmtop and inpcrd files.
    3) section3.htm - Minimising the structure.
    4) section4.htm - Heating the system up.
    5) section5.htm - Production MD.
    6) section6.htm - Analysing the Results
    7) section7.htm - Testing the Stability of the NMR structure.

WARNING: some of the calculations in this tutorial can take a very long time to run. I used 16 cpus of a 1.3GHz shared memory SGI Altix to run these calculations and they took a total of 27 hours to run. While I urge you to run these simulations in order to familiarise yourself with SANDER I provide the relevant output files so that you can still follow the tutorial even if you don't have sufficient computing power at your disposal.

WARNING 2: There is no guarantee that if you run these simulations yourself that you will get the exact same answers as I. Difference in machine architecture leads to rounding errors in the calculation which result in different simulations on different machines exploring different regions of phase space. The average properties, however, should be comparable. As you will see though as you progress through this tutorial there is no guarantee that you can reproduce the results described in the paper. Indeed, even though we setup what should be an almost identical simulation to the paper we fail to achieve the results they publish. This is likely because our simulations are not long enough or have been trapped in local minima. Nevertheless this tutorial shows some interesting behaviour.

This paper describes the simulation of peptide folding using an all atom classical simulation and a slightly modified version of the AMBER FF99 force field. "Trpcage" is a 20 residue amino acid sequence that has been optimised by the Andersen group at the University of Washington. It is currently the smallest protein to display two state folding properties and is stable at room temperature. The small size of this protein makes it an ideal candidate for computational folding simulations. When the original folding simulations were done the experimental structure had not been solved and so the prediction was made without reference to experiment. When the experimental structure was solved the predicted structure was within 1.4 angstroms RMSD. This is exceptionally good for a simulation started from a straight chain linear sequence with no restraints.

In this tutorial we will attempt to reproduce the results obtained in this paper. The simulation protocol used will be very similar although due to time restraints we will run only one simulation of approximately 50 ns in length. This should be sufficient to reproduce the folding results although if this were new work we would have to extend our simulation as well as carry out other similar simulations to verify our results and the stability of the predicted structure. A word of warning here, due to the length of these simulations the results obtained on different machines and/or different numbers of processors will be different. This is due to the way molecular dynamics works, small variations in the order of execution and rounding in the floating point calculations mean that the trajectories sampled by different machines will diverge over time. This is not an error or a bug, neither is one simulation more correct than another. It is simply that the two simulations are exploring different regions of phase space. If we average the results, and we have run long enough simulations, the two machines should give the same result, they just arrive at it via a slightly different path. Hence it is difficult for us to exactly reproduce the results of the paper in this tutorial but we can try to re-create the important result, that it is possible to predict the folded structure of a 20 amino acid peptide using AMBER.