QSAR Studies Of Organic Compounds By Optimized Radial Basis Function Network

QSAR (quantitative structure-activity relationship, QSAR) study is relatively active in the international research field. It was widely used in the fields of biology, medicine, chemistry, agriculture, and environment. The construction of QSAR model is the key step of QSAR studies. Radial basis function network as an important non-parameter modeling tool has been proven fruitful in quantitative structure-activity relationship (QSAR) studies. It may be due to the fact that RBFN holds great potential in approximating various nonlinear relationships between the descriptors and bioactivities within sufficient accuracy. This is also the right advantage of RBFN over other methods, e.g., partial least-squares (PLS) regression. The issues of overfitting and local optima often happened during RBFN training. In this paper, to rectify this situation, two researches were done:(1) Regression tree (RT), allied with hybrid particle swarm optimization (PSO) algorithm, were invoked to configure an RBFN to form the HPSORTRBFN algorithm in the present study. Discrete PSO was invoked to obtain an RT of the right size, the regions in the instance space defined by the leaf nodes of the grown RT were transformed into the centers in RBF units and the number of leaf nodes acted as the network structure. The splitting variables in RT became the inputs in RBFN. The widths and weights in RBFN were simultaneously optimized by continuous PSO. HPSORTRBFN was applied to predict the anti-HIV activities of1-[(2-Hydroxyethoxy) methyl]-6-(phenylthio) thymine (HEPT) analogues and the bioactivities of flavonoid derivatives. The results showed RT allied with HPSO is able to configure a globally optimal RBFN and HPSORTRBFN owns superior modeling performance to RBFN and RT.(2) In this chapter, the performance of RBFN was improved from another aspect, named PSORBFPLS. In PSORBFPLS, continuous particle swarm optimization algorithm (PSO) was used to build radial basis function network (RBFN) based on partial least squares (PLS). This method firstly transforms the descriptors into the hidden layer output by using radial basis function, and then, PLS is used to link the output of the hidden layer and the bioactivities of the samples. In order to adaptive adjustment of the nonlinear transformation of the original descriptors, PSO was invoked to optimize the centers and radius involved in RBFN. The number of latent variables involved in PLS to link the output of the hidden layer and the bioactivities was automatically determined by the F-statistics. Two QSAR data sets were employed to validate the performance of the proposed algorithm. The results showed that the new method can effectively improve the performance of RBFN, converging quickly to the optimal solution and improving the generalization ability of RBFN in a great extent.