Construction of an optimal immune network model based on the modified swarm algorithm

Abstract

the approach of artificial immune systems, for the computer molecular design of new drugs and
prediction the "structure-property/activity"relationship (QSAR) of chemical compounds is an
actual problem. The article is devoted to the solution of the task of QSAR on the construction of
immune network model based on the choice of optimal set of descriptors to facilitate the selection of
new chemical compounds for candidate drugs with predefined properties. According to the concept
of multialgorithmic approach development of optimal immune network model and allocation of
informative descriptors is carried out on the basis of swarm intelligence algorithms. In this work
comparison of standard particle swarm optimization algorithm (PSO) and modified inertia weight
particle swarm optimization (IWPSO) is described for selection of informative descriptors on the
example of drug compounds of the sulfanilamide group with various pharmacological activities. The
choice of the parameters (fitness functions, population size, the number of iterations, etc.), which
define performance of the offered algorithms for creation of optimal set of descriptors is analysed.
The results of modeling of dependence of fitness function values on the number of iterations in
software products WEKA and Yarpiz (PSO) are given.