Chemoinformatic Prediction Of Bacterial Metabolite Concentration And Its Applications In Antibacterial Discovery

With the development of omics, such as genomes, transcriptomes, and protomes, life science has entered the era of omics. Metabolomics is a subject that analyzes all pathways and metabolites in a high-throughput manner, which is another important research field in system biology after genomics, transcriptomics and proteomics. Metabolites are the end products of the cellular control process and the research object of metabolomics. The types and concentration changes comprise the final response of biological systems for genetic variation or disturbance in the environment. Therefore, the metabolome is closer to the biological phenotype than the genome and protome. Metabolomic research focuses on the qualitative and quantitative analyses of metabolite types and concentrations, as well as the reconstruction or simulation on metabolite networks. In the past, the metabolomic studies were limited to high-throughput experimental technologies. However, metabolite network analyses have rapidly developed based on the bioinformatics tools and have provided strong technical support for the discovery of antimicrobial drug targets.First, we collected metabolites of E.coli from the KEGG database, reconstructed metabolic networks, and computed the physico-chemical properties. By correlation analysis, we determined that metabolite concentrations and their physico-chemical properties (i.e., polarity, molecular weight, and solubility) have a strong correlation. The linear prediction model and support vector regression (SVR) model were constructed based on networks degree and84physico-chemical properties. We obtained the linear regression equation through stepwise regression. Ten physico-chemical properties and the degree were screened by recursive feature elimination (RFE) method. Compared with the linear regression and network thermodynamic methods, the prediction results of the SVR model with RFE were more accurate, consistent with our expectations. In particular, the metabolite concentrations of E.coli are related to the metabolic network degree, polarity, and water solubility. Second, we applied the SVR model of E.coli to animal and plant pathogens. A concentration prediction website was built based on the10physico-chemical properties and network degree. The prediction concentration of10bacteria types, such as Pectobacterium carotovorum, Staphylococcus aureus, and M. tuberculosis can be queried from this website. Other bacteria metabolite concentrations can also be predicted on this website.Third, we proposed a useful criterion for selecting antimicrobial targets based on metabolite concentration. Use this criterion, we clearly explained previous studies of interesting phenomena.Finally, we identified three flavonoid targets by metabolite concentrations. Docking results reveal that substitution of galloyl or glycosides at position3of the heterocyclic pyrane ring in flavonoids enhances the binding affinity to three targets(i.e., FrdA, PyrD, and FabI). The biological functions of the three targets are consistent with the reported antibacterial mechanisms of flavonoids. Based on the combined metabolite concentration discrimination criterion, the water solubility of the18flavonoids was generally higher than the metabolite substrates concentration. This result shows that flavonoids have good antibacterial activity. Thus, the three targets can be potential antibacterial targets.Our study provided a new idea and criterion for studying target screening at the metabolic level, which were successfully applied in screening flavonoid targets.