Prediction of time series microarray data using neurofuzzy networks

Abstract

There have been many studies recently that predict the interactions between genes and reconstruct the gene control network. In this paper, we propose the approach to predict the expression values between the genes of the yeast cell using a neural network based on weighted fuzzy membership function. This neuro fuzzy system makes the exact prediction possible through choosing best rules automatically. Features extracted from original data are used for learning. We extract the five features and they take into account the characteristics of time series by using wavelet transform, Current Position (CP) and time point. The best features to be good for prediction are selected through the Bounded Sum Weight of the weighted fuzzy membership function. The selected features are defuzzified through the Takagi-Sugeno method to calculate the prediction values of original gene expression data. We evaluate mean square error to indicate prediction accuracy of the proposed approach and then compare to the existing algorithm RNN using the neural network. The proposed method outperformed RNN.