Abstract
Composition-Activity Relationship (CAR) modeling is a novel approach to evaluate the quality and identify active components of herbal medicine. In this study, Grid Search Method (GSM) and Heuristics algorithms, particularly Genetic Algorithm (GA) and Particle Swarm Optimization (PSO), were adopted to determine the optimal parameters automatically. Then, support vector regression (SVR) combined with a linear kernel function or a radial basis kernel function (RBF) and back propagation artificial neural networks (BPANN) were employed to construct the model that correlated the main chemical components with the cytotoxicity of the essential oil from Curcuma longa L., respectively. Considering the robustness and predictive ability, the ν-SVR-RBF-PSO model had the best performance in various tests performed in this paper. Nine components were then identified to have significant cytotoxicity based on the superior model and Mean Impact Value (MIV) analysis. An optimal model can therefore be a useful tool to predict the bioactivity for quality evaluation and active components identification of herbal medicine.
Acknowledgments
The authors are grateful for the financial support provided by the National Natural Science Foundation of China (No. 81102900) and the help of Associate Professor Zhang Han from the Department of Automation and Intelligence Science, Nankai University.
Notes
a ERGS: ϵ-SVR-RBF-grid search, ERGA: ϵ-SVR-RBF-GA, ERPS: ϵ-SVR-RBF-PSO; ELGS: ϵ-linear kenel function-grid search; ELGA: ϵ-SVR-linear kenel function –GA; ELPS: ϵ-SVR-linear kenel function-PSO; NRGS: v-SVR-RBF-grid search; NRGA: v-SVR-RBF-GA; NRPS: v-SVR-RBF-PSO; NLGS: v-SVR-linear kenel function-grid search; NLGA: v-SVR-linear kenel function-GA; NLPS: v-SVR-linear kenel function-PSO.
b MSE: mean square error, RSE: relative standard error, R: correlation coefficient of training phase, Q: correlation coefficient of testing phase.
a UN = unidentified.
The authors declare no conflict of interest.