Response Surfaces Method and Artificial Intelligence Approaches for Modeling the Effects of Environmental Factors on Chlorophyll a in Isochrysis galbana

This study reported the condition optimization for chlorophyll a (Chl a) from the microalga Isochrysis galbana. The key parameters affecting the Chl a content of I. galbana were determined by a single-factor optimization experiment. Then the individual and interaction of three factors, including salinity, pH and nitrogen concentration, was optimized by using the method of Box-Benhnken Design. The highest Chl a content (0.51 mg/L) was obtained under the optimum conditions of salinity 30% and nitrogen concentration of 72.1 mg/L at pH 8.0. The estimation models of Chl a content based on the response surfaces method (RSM) and three different artificial intelligence models of artificial neural network (ANN), support vector machine (SVM) and radial basis function neural network (RBFNN), were established, respectively. The fitting model was evaluated by using statistical analysis parameters. The high accuracy of prediction was achieved on the ANN, SVM and RBFNN models with correlation coefficients (R-2) of 0.9113, 0.9127, and 0.9185, respectively. The performance of these artificial intelligence models depicted better prediction capability than the RSM model for anticipating all the responses. Further experimental results suggested that the proposed SVM and RBFNN model are efficient techniques for accurately fitting the Chl a content of I. galbana and will be helpful in validating future experimental work on the Chl a content by computational intelligence approach.

Automated summary

This study optimized the conditions for chlorophyll a (Chl a) extraction from the microalga Isochrysis galbana. The key parameters affecting Chl a content were determined through a single-factor optimization experiment, and then the interaction of three factors (salinity, pH, and nitrogen concentration) was optimized using the Box-Benhnken Design. The artificial intelligence models (ANN, SVM, and RBFNN) showed better prediction capability for Chl a content than the traditional response surfaces method (RSM).