Reductive and Reproductive Neural Networks Predict Enzyme Classification at the Time-of-Reduction-Recovery

We propose a neural network model, having a reductive and reproductive topology to find the optimal predictive timing and then demonstrate to improve enzyme functional classification. The model is similar to the ordinary multi-layered network, except that it has rules to decrease and increase the hidden layer units according to the evaluated network performance during the predictive procedure. Various types of the topology change are tested in the enzyme functional classification, which is a computational approach based on the amino acid sequences. A preferable timing for the prediction is found at an appropriate iteration time after the unit reduction, at a "time-of-reduction-recovery," providing predictive rate improving upon that brought about the fixed topology of various sizes. The model, having everlasting predictability, will be of help in developing a system with fast and high predictive performance, though some extra iterations are required depending on the application. Key-Words: Reductive/reproductive neural networks, Variable topology of neural networks, Optimal predictive timing, Time-of-reduction-recovery prediction, Enzyme functional classification