Authors: Shaomin Yan, Guang Wu
The working conditions for enzymatic reaction are elegant, but not many optimal conditions are documented in literatures. For newly mutated and newly found enzymes, the optimal working conditions can only be extrapolated from our previous experience. Therefore a question raised here is whether we can use the knowledge on enzyme structure to predict the optimal working conditions. Although working conditions for enzymes can be easily measured in experiments, the predictions of working conditions for enzymes are still important because they can minimize the experimental cost and time. In this study, we develop a 20-1 feedforward backpropagation neural network with information on amino acid sequence to predict the pH optimum for the activity of beta-glucosidase, because this enzyme has drawn much attention for its role in bio-fuel industries. Among 25 features of amino acids being screened, the results show that 11 features can be used as predictors in this model and the amino-acid distribution probability is the best in predicting the pH optimum for the activity of beta-glucosidases. Our study paves the way for predicting the optimal working conditions of enzymes based on the amino-acid features.
See also: Comments to Paper