NSSGRN: Network Structure Selection Method for Gene Regulatory Network Construction.

Gene regulatory networks (GRNs) play essential roles in cellular control and various biological processes. Analyzing gene expression data and inferring GRNs provides crucial insights into organismal growth, development, and disease mechanisms. However, prevailing inference approaches often concentrate on a limited gene expression feature set and tend to analyze network structure from a single perspective, thus restricting a comprehensive understanding of gene relationships. To address this issue, we introduce a novel network structure selection method for GRN construction (NSSGRN), considers the isomorphism and complementarity of network structures generated by several classical methods, and integrates them to infer the network structure. Specifically, NSSGRN firstly generates an initial gene relationship prioritization from knockout data. Second, several methods are integrated by considering the isomorphism and complementarity of their results. Finally, the integrated network structure is optimized by a scoring based method to increase true positives and reduce false positives. Experiments on two challenging datasets (25 networks in total) shows that NSSGRN outperforms nine other advanced methods in overall performance, demonstrating its effectiveness in enhancing the accuracy of GRN construction. The code is available at https://github.com/Xtu-LWGroup/NSSGRN.git.