Meld: Modeling Peptide-Protein Interactions

Abstract

We have recently developed Meld, a framework to Model with limited data. This framework combines sparse information coming from different experimental, bioinformatics and even evolution sources into a physics based methodology. The physics are implemented through an atomistic force field. An improved version of a hybrid Hamiltonian/temperature replica exchange [1] procedure allows us to handle the limited data along with the physics in a fashion that obeys detail balance. After our initial use in the CASP experiment (critical assessment of structure prediction) to predict the structures of proteins, we are now interested in showing how this procedure can help us to correctly model cases in which small peptides are interacting with proteins. This procedure has the advantage of having both peptide and protein as flexible units. Input data is introduced in the calculation as restraints. A naïve approach in which all the restraints are imposed at the same time would be a failure. Therefore, we allow for errors in the data by enforcing only a fraction of those restraints. We allow the physics to decide which restraints are the most compatible with the system. In this way, the conformational space is greatly reduced, and in combination with 100x improvements in sampling efficiency coming from GPU, allows us to converge into possible solutions. After our calculation is done, clustering methods allow us to identify the candidate docking regions. We plan to combine the top clustering solutions with all atom free energy methods based on confinement techniques [2] to identify the most likely binding site. 1. Sugita, Y., A. Kitao, and Y. Okamoto, “Multidimensional replica-exchange method for free-energy calculations”. Journal of Chemical Physics, 113(15): p. 6042-6051 (2000). 2. Tyka, M., Clarke, R. and Sessions, R. An efficient, path-independent method for free-energy calculation, J. Phys. Chem. B,110, p. 17212-17220 (2006).