Abstract:

Back-propagation artificial neural network combined with genetic algorithm(BPANN-GA) was applied to optimize the high performance liquid chromatography(HPLC) conditions for the determination of epothilone A(Epo A) and epotholine B(Epo B) simultaneously in Polyangium cellulosun metabolites. The concentration of acetonitrile in mobile phase, column temperature and the pH of mobile phase were selected as casual factors and a three-factor-twelve-level uniform design(U₁₂³) was used for experiment design. A two-layer back-propagation artificial neutral network(BPANN) was applied to model for the correlation between the casual factors and chromatography optimization function(COF) values, which was chosen as the criterion. Levenberg-Marquardt algorithm was used for training the BPANN. The BPANN model was optimized by selecting the most suitable numbers of hidden neurons depending on the degree of approximation(D_a). The correlation coefficient(R) between the COF values obtained by BPANN model and the experiment values was 0.98165. While the optimum BPANN model was developed, genetic algorithm(GA) was applied to find out global dissolution in modeling range. The optimum HPLC conditions obtained by BPANN-GA were as follows: the concentration of acetonitrile in mobile phase was 29.2%(volume fraction); the column temperature was 34 ℃ and pH of mobile phase was 4.23. The validation experiment at the optimum conditions was performed, and the satisfied chromatogram was obtained.

Key words: Back-propagation artificial neural network(BPANN); Genetic algorithm(GA); Epothilone; High performance liquid chromatography(HPLC)