Extraction of pepsin-soluble collagen from grass carp (Ctenopharyngodon idella) skin using an artificial neural network

A multilayer feed-forward neural network trained with an error back-propagation algorithm was used to evaluate the effects of pepsin amount, reaction time and pH on the yield of pepsin-soluble collagen. A positive correlation was observed between the yield and the amount of pepsin and also the reaction time. The yield increased with an increase of pH to nearly 3, thereafter yield decreased. The trained network gave a regression coefficient (r²) of 0.97 and a mean squared error (MSE) of 0.21, which implied a good generalisation of the network. Based on the genetic algorithm, the optimal extraction conditions to obtain the highest yield were determined to be pH 3.4, 53.3 unit/mg of pepsin and 35.2 h. The predicted yield value was 30.3%. As the estimated optimal extraction conditions were used in the actual preparation of the pepsin-soluble collagen, the yield was measured experimentally to be 29.3 ± 0.8%, which was not significantly different (p > 0.05) from the predicted value. The response surface plots showed the yield of pepsin-soluble collagen as a function of two factors under various extraction conditions.