Computational Drug Discovery and Design

graphics-cards/gtx-titan-black). The MD procedures outlined

below have been developed with the Amber 14 and AmberTools

13 suite of programs [45] but may work with other versions of the

software. Some of the data conversion was done with the Open-

Babel program (http://openbabel.org/wiki/Main_Page). The

docking has been performed with the AutoDock Vina program

[46] using the AutoDock Tools [47] as well as with the OpenEye

software suite (OpenEye Scientific Software, Santa Fe, NM,http://

http://www.eyesopen.com). Chemical structures were drawn with Mar-

vinSketch from ChemAxon (Marvin 6.1.5, 2013,http://www.

chemaxon.com). Visualization has been done with Pymol v.1.8

(https://www.pymol.org) and Vida v.4.3 (OpenEye).

3 Methods

The objective of this chapter is to describe in detail how to perform

virtual screening preserving the flexibility of both the receptor and

the ligand. The receptor’s flexibility is mimicked by docking to a

representative set of receptor’s structures, which can be obtained

from the clustering analysis of an MD trajectory. The ligand’s

flexibility depends on the number of the rotatable bonds. This

problem is routinely handled by the recent software. The MD

runs are executed on GPU cards rather than on CPUs in order to

profit from the acceleration of the calculation offered by hugely

increased numbers of processing units present in GPUs. Details of

the procedures which refer specifically to GPU computing will be

mentioned explicitly in the text below.

3.1 Preparation
of the Receptor’s Input
Files

1. Obtain the PDB file of the protein you are interested
   in. Typically, it will come from the PDB database (http://
   http://www.rcsb.org, http://www.pdb.org/)) if the structure was
   experimentally determined using X-ray, NMR, or cryo-EM
   studies, or it can be obtained by homology modeling
   [48–50]. The NMR files include protons, the others usually
   only heavy atoms (seeNote 1).


2. Add missing residues. Unstructured loops and other flexible
   regions of the receptor may not appear in crystallographic PDB
   files. The MD software, however, needs a complete molecule.
   We use mainly MODELLER v. 9 [48] to fill in the structural
   gaps. The python script “loop.py”, supplied with examples of
   MODELLER’s use, can be used for this purpose. Alter the line:

self.residue_range(’X:A’, ’Y:A’)

to define your own region to model, substituting X, Y, and A

by the relevant residue range and chain symbol, respectively.

Molecular Dynamics in Virtual Screening 149