Chemistry Reference
In-Depth Information
absence of accompanying chemical synthetic routes. It may be more useful to
generate restricted
de novo
libraries for specific applications. It may also be of
interest to generate web servers that allow fast computational generation of
de
novo
libraries and estimate the inherent difficulties of synthetizing chemical
compound and their synthetic accessibility.
In the last decades, chemical libraries have provided valuable resources for drug
design, ready-to-use tools, reliable and comprehensible selection of compounds
from free and commercial sources that are purchasable for further wet lab
experiments.
From a computational point of view, a fine-grained model divides the populations
into several subpopulations, each assigned to different processors or machines.
Each population can only exchange information with other connected
subpopulations. The master-slave model (genetic algorithm) assigns the
populations to several subpopulations. These are subsequently assigned to
different processors in order to enhance the computer speed. The coarse-grained
model uses the concept of migration and island in order to exchange information
between subpopulations.
The parallel computation models take into account both the computation resource
and the architecture of the search algorithm. The focus is on how the
chromosomes
exchange
and
communicate
information
between
the
subpopulations in order to affect the search population.
Considerable effort is being investing in high performance computing in order to
speed up the protein-ligand docking prediction search process as well as sampling
methods to provide better initial seeds and end results. Cloud computing is a
promising research area for docking predictions by enhancing the tremendous
amount of computing and storage resources whereas the focus is on how to divide
and dispatch search algorithm computations to computer nodes to improve
performance.
An important problem for heuristic algorithms is the quality of solution and the
balance of computation cost. Most computing nodes are not centralized for grid
