Adaptive Elman Model of Gene Regulation Network Based on Time Series Data

Background: Time series expression data of genes contain relations among different genes, which are difficult to model precisely. Slime-forming bacteria is one of the three major harmful bacteria types in industrial circulating cooling water systems. Objective: This study aimed at constructing gene regulation network(GRN) for slime-forming bacteria to understand the microbial fouling mechanism. Methods: For this purpose, an Adaptive Elman Neural Network (AENN) to reveal the relationships among genes using gene expression time series is proposed. The parameters of Elman neural network were optimized adaptively by a Genetic Algorithm (GA). And a Pearson correlation analysis is applied to discover the relationships among genes. In addition, the gene expression data of slime-forming bacteria by transcriptome gene sequencing was presented. Results: To evaluate our proposed method, we compared several alternative data-driven approaches, including a Neural Fuzzy Recurrent Network (NFRN), a basic Elman Neural Network (ENN), and an ensemble network. The experimental results of simulated and real datasets demonstrate that the proposed approach has a promising performance for modeling Gene Regulation Networks (GRNs). We also applied the proposed method for the GRN construction of slime-forming bacteria and at last a GRN for 6 genes was constructed. Conclusion: The proposed GRN construction method can effectively extract the regulations among genes. This is also the first report to construct the GRN for slime-forming bacteria.