EE Seminar: Generating funnel graphs efficiently using GPU accelerated docking

~~
Speaker: Michael Zabejansky
M.Sc. student under the supervision of Prof. Haim Wolfson and Prof. Boaz Pat-Shamir

Wednesday, July 27th, 2016 at 15:30
Room 011, Kitot Bldg., Faculty of Engineering

Generating funnel graphs efficiently using GPU accelerated docking

Abstract

Protein docking is commonly described as a search for the minimum binding energy conformation of a complex. Sampling multiple receptor-ligand conformation binding energies in the surrounding of the conformation can result in an ensemble of low-energy conformations, which will form a funnel like energy landscape in the energy-rmsd graph.
The energy landscape in a graph can inform us on different low energy conformations. Generating a dense binding energy graph can provide a powerful tool not only for validating prediction solutions, but also for searching better complex conformations around. However, generating such a graph demands sampling an accurate energy function thousands of times, which is computationally demanding. This task can take days with currently available algorithms.
The task of performing multiple independent calculations fits perfectly for Graphic Processing Units (GPU) hardware. We developed a dedicated algorithm for generating multiple samples of a complex binding energy, which exploits the power of GPUs to provide input for funnel graphs.
In this talk, I will present our algorithm, which accelerates the FiberDock docking algorithm using GPUs and parallelization in order to perform multiple calculations of the binding score fast and efficiently. The algorithm provides a useful tool for generating funnel graphs for different needs.