The archive containing the scripts presented in this chapter is
available for download from the following URL:http://cribligand.
ipbs.fr/cheminfo-scripts.html.4 Notes
- PDB files from NMR studies usually contain 20 superposed
structures. Pick the first one, as it should have the least con-
straint violations. The choice is not critical because the MD will
relax the molecule and drive it to equilibrium. - The loop modeling as described above works up to the loop
length of 30 residues. For longer loops a different strategy has
to be adopted. One can model loops incrementally, a fragment
at a time. Another possibility is to run an MD simulation with
the loop as a separate chain, with distance constraints on loop
termini to obtain spatial proximity to the receptor, then run-
ning the MD again to equilibrate the loop incorporated in the
structure. - If you want to simulate deprotonated HIS,seeref. 71.
- tleap renumbers residues beginning from 1. Subsequent resi-
due referencing (e.g., for the “bond” command) must take this
into account. - If multiple GPU cards are supposed to execute the MD proto-
col, the first two stages (i.e., the two consecutive minimiza-
tions) have to be performed on a single card. At the time of the
writing this part of the code is not yet parallelized for GPUs.
Multiple GPU cards can work on the MD protocol beginning
with the third stage, i.e., from the short MD run with sample
heating. - In order to use GPUs for MD, install the appropriate version of
NVidia’s CUDA library on your system, then compile the
CUDA version of the software (e.g., Amber’s pmemd.cuda).
This will create a binary compatible with your hardware. There
is a problem in NVidia’s software with the detection of GPU
cards and the assignment of unique identifiers. The order in
which GPU cards are detected is different depending on
whether we launch a command from a terminal (nvidia-smi),
from a GUI interface (nvidia-settings) or via the CUDA library
(deviceQuery). In each case the identifiers of GPU cards may
be different, which is confusing. However, in practice only the
output from deviceQuery is valid for the launch of the GPU
software. - The kclust program handles up to 50000 files. In practice, this
is not a significant limitation. However, its output contains an
error: the PDB files written by kclust show residue numbers in
Molecular Dynamics in Virtual Screening 173