A multi-objective evolutionary approach to reconstruct gene regulatory network using recurrent neural network model

With the advent of various data assaying techniques, gene expression time series data have become a useful resource to investigate the complex interactions occurring amongst the transcription factors and genes. While a number of methodologies have been developed to describe Gene Regulatory Network (GRN), the presence of high noise in gene expression data have made the estimation of non-linear interactions among the genes an ill-posed one. In this work, a multi-objective evolutionary strategy has been proposed to efficiently reconstruct the skeletal structure of the biomolecular network using the Recurrent Neural Network (RNN) formalism. Moreover, this work presents a second criterion for model evaluation to exploit the sparse and scale free nature of GRN. This evaluation criterion systematically adapts the max-min in-degrees to effectively narrow down the search space, which reduces the computation time significantly and improves the model accuracy. The two well-known performance measures applied to the experimental studies on synthetic network with expression data having different noise-levels. The experimental results clearly demonstrate the suitability of the proposed method in capturing gene interactions correctly with high precision even with noisy time-series data. The experiments carried out on analyzing well-known real expression data set of the SOS DNA repair system in Escherichia coli show a significant improvement in reconstructing the network of key regulatory genes.