A Novel Technique for Reduction of False Positives in Predicted Gene Regulatory Networks

TL;DR: Experimental results suggest that the proposed methodology is capable of reducing the number of false positives, thus, increasing the overall accuracy and the biological plausibility of the predicted genetic networks.

Abstract: In this paper, we have proposed a novel method for the reduction of the number of inferred false positives in gene regulatory networks, constructed from time-series microarray genetic expression datasets. We have implemented a hybrid statistical/swarm intelligence technique for the purpose of reverse engineering genetic networks from temporal expression data. The theory of combination has been used to reduce the search space of network topologies effectively. Recurrent neural networks have been employed to obtain the underlying dynamics of the expression data accurately. Two swarm intelligence techniques, namely, Particle Swarm Optimisation and a Bat Algorithm inspired variant of the same, have been used to train the corresponding model parameters. Subsequently, we have identified and used their common portions to construct a final network where the incorrect predictions have been filtered out. We have done preliminary investigations on experimental (in vivo) data sets of the real-world SOS DNA repair network in Escherichia coli. Furthermore, we have implemented our proposed algorithm on medium-scale networks, consisting of 10 and 20 genes. Experimental results are quite encouraging, and they suggest that the proposed methodology is capable of reducing the number of false positives, thus, increasing the overall accuracy and the biological plausibility of the predicted genetic networks.