Predicting time sequences in gene expression using dynamic neuro-fuzzy networks

Gene regulation at the cellular level is a dynamic process; and neural networks with the capability of being trained and regulated based on training data, and fuzzy systems with their interpretability, are suitable algorithms for this type of computations. In this research, methods based on neuro-fuzzy networks were employed for predicting the complicated relationships that exist among genes. In the proposed method, genes that have the greatest effect on each other are found and considered as regulatory genes. Then the types of their relationships including those with inhibitory, activating, and neutralizing effects are determined, and the gene regulatory network is mapped. The standard microarray dataset related to 12 typical genes that influence the germination cycle of the yeast species Saccharomyces cerevisiae was used for training. Obtained results were validated using previous biological laboratory results. It was found that the proposed method reduced the number of extracted rules for input-output space fragmentation by 15%, which substantially reduced the required computations, while the sum square error of the algorithm declined by -0.3 compared to the method that resembled the proposed algorithm the most.