Development of docking programs for Lomonosov supercomputer

Abstract

The initial step of the rational drug design pipeline extremely needs an increase ineffectiveness. This can be done using molecular modeling: docking and molecular dynamics. Dockingprograms are popular now due to their simple idea, quickness and ease of use. Nevertheless accuracy ofthese programs still leaves much to be desired and discovery by chance and experimental screening stillplay an important role. Docking performs ligand positioning in the target protein and estimates theprotein-ligand binding free energy. While in many cases positioning accuracy of docking is satisfactory,the accuracy of binding energy calculations is insufficient to perform the hit-to-lead optimization. Theaccuracy depends on many approximations which are built into the respective model. We show that allsimplifications restricting docking accuracy can be withdrawn and this can be done on the basis ofmodern supercomputer facilities allowing to perform docking of one ligand using many thousandcomputing cores. We describe in short the SOL docking program which is used during years for virtualscreening of large ligand databases using supercomputer resources of Lomonosov Moscow StateUniversity. SOL to some extent is organized similarly to popular docking programs and reflects theirlimitations and advantages. We present our supercomputer docking programs, FLM and SOL-P, developedover the past 5 years for Lomonosov supercomputer of Moscow State University. These programs arefree of most important simplifications and their performance shows the road map of the docking accuracyimprovement. Some results of their performance for very flexible ligand docking into the rigid protein anddocking of flexible ligands into the protein with some moveable protein atoms are presented. The socalled quasi-docking approach combining a force field and quantum chemical methods is described and itis shown that best docking accuracy is reached with the PM7 method and the COSMO solvent model.