Gene Regulatory Network Inference Relating to Glycolysis in Escherichia coli with Causal Discovery Method Based on Machine Learning

Escherichia coli LS5218 is an attractive host for producing polyhydroxybutyrate. The strain, however, strongly requires heterologous gene expressions like phaC for efficient production. For enhancing the production, the whole gene expressions relating to end product-producing flow should be optimized so that not only heterologous induced-genes but also other relating genes are comprehensively analyzed on the transcription levels, resulting in normally time-consuming mutant-creation. Additionally, the explanation for each transcriptional relationship is likely to follow the relationships on known metabolic pathway map to limit the consideration. This study aimed to infer gene regulatory networks within glycolysis, a central metabolic pathway in LS5218, using machine learning-based causal discovery methods. To construct a directed acyclic graph representing the gene regulatory network, we employed the NOTEARS algorithm (Non-combinatorial Optimization via Trace Exponential and Augmented lagRangian for Structure learning). Using transcription data of 264 time-resolved sampling points, we inferred the gene regulatory network and identified several distal regulatory relationships. Notably, gapA, a key enzyme controlling the transition between the preparatory and rewarding phases in glycolysis, was found to influence pgi, the enzyme at the pathway’s entry point. These findings suggest that inferring such nonlocal regulatory interactions can provide valuable insights for guiding genetic engineering strategies.