Artificial Immune Network For Parameter Optimization In Pharmacokinetics

Artificial immune network is a key branch of artificial immune system, which is based on the clonal selection and immune network principles of the immunology. In this paper, new methods of artificial immune network for parameter optimization were investigated to optimize pharmacokinetic parameters. Then the hybrid algorithms of artificial immune network and simplex were discussed in order to improve optimization performance. The main work reported in this paper can be concluded as follows:First, the PKAIN (PharmacoKinetics Artificial Immune Network) artificial immune networkwas experimentally tested with several typical compartment models and compared with genetic, simplex and lest squares algorithms. In order to solve nonlinear pharmacokinetic models, the iterative inverse Laplace Transform method was adopted to obtain numerical solution of the predicted models with a given set of parameters. Then the PKAIN was extended by combing with numerical solutions to search for optimal parameters of the nonlinear pharmacokinetic models. Since the iterative inverse Laplace Transform method support distribution computing, the PKAIN can be parallelized to improve its efficiency. Moreover, Normalized network suppression function was proposed and compared with current network suppression methods. Experimental results showed that the normalized network suppression function is able to simplify parameter setting of network suppression procedure of the artificial immune network.Second, a variety of hybrid methods which are combined with artificial immune network and simplex were mainly investigated. An iterative partition-based concurrent simplex operator was proposed at first. The artificial immune network with this operator denoted as PKAIN2 was adopted to improve local searching ability of the original PKAIN algorithm. Experimental results indicated that the PKAIN outperformed the PKAIN algorithm in the term of optimization ability. However, time consumption is increased at the same time.Third, in order to improve searching speed, a hybrid algorithm serializing the PKAIN and the simplex method, named as PKAIN_spx was presented. The artificial immune network was used to do coarse-grained search. After convergence, the PKAIN sent several better solutions to the simplex method to do finer search simultaneously. Experiments about several bench functions demonstrated that the PKAIN_spx outperformed the PKAIN and the PKAIN2 artificial immune networks in term of running speed. Given a certain number of initial solutions which are obtained by the artificial immune network, the PKAIN_spx algorithm is able to find global optimal solutions at the same time.