Section: New Results

Pepsi-Piper: rigid docking predictions using Pepsi potentials into Piper code

Participants : Sergei Grudinin, Emilie Neveu, Dima Kozakov, Dzmitry Podgorny.

This work is the continuation of the Pepsi-Dock project that aims to develop fast predictions of putative docking poses using accurate knowledge-based potentials functions to describe interactions between proteins. The goal is to integrate the precise, and yet easy to compute, distance-based pairwise knowledge-based potentials [58] into the Piper search code [48] in order to compare its exhaustive search with the Hex one. The former samples the conformations using a cartesian grid while the latter, a spherical one. We proved our potential used in Hex can predict the structures of complexes with a really good success rate, the main limitation being the lack of precision of the spherical sampling when the separation distance of the two proteins is too large. We think predicting docking combining our potential and a sampling search based on a cartesian grid as in Piper will achieve greater results, but will require more computational time.

We first adapt our potential to the Piper code and showed that the ranking results on the data set used for training are better than the ranking provided by Piper [25] : when the potential is used to sort the conformations, the correct solution is found in the first ten for 85% cases, while Piper found it in only 25% cases. The next step is to use the cartesian sampling to make docking predictions. When the Piper code will be ready to integrate our potential, we will be able to confront with other knowledge-based potentials such as the one initially used in Piper, DARS.