Background: Healthcare-associated infections (HAIs) are a serious public health problem. They can be associated with morbidity and mortality and are responsible for the increase in patient hospitalization. Antimicrobial resistance among pathogens causing HAI has increased at alarming levels. In this paper, a robust method for analyzing genome-scale metabolic networks of bacteria is proposed in order to identify potential therapeutic targets, along with its corresponding web implementation, dubbed FindTargetsWEB. The proposed method assumes that every metabolic network presents fragile genes whose blockade will impair one or more metabolic functions, such as biomass accumulation. FindTargetsWEB automates the process of identification of such fragile genes using flux balance analysis (FBA), flux variability analysis (FVA), extended Systems Biology Markup Language (SBML) file parsing, and queries to three public repositories, i.e., KEGG, UniProt, and DrugBank. The web application was developed in Python using COBRApy and Django. Results: The proposed method was demonstrated to be robust enough to process even non-curated, incomplete, or imprecise metabolic networks, in addition to integrated host-pathogen models. A list of potential therapeutic targets and their putative inhibitors was generated as a result of the analysis of Pseudomonas aeruginosa metabolic networks available in the literature and a curated version of the metabolic network of a multidrug-resistant P. aeruginosa strain belonging to a clone endemic in Brazil (P. aeruginosa ST277). Genome-scale metabolic networks of other gram-positive and gram-negative bacteria, such as Staphylococcus aureus, Klebsiella pneumoniae, and Haemophilus influenzae, were also analyzed using FindTargetsWEB. Multiple potential targets have been found using the proposed method in all metabolic networks, including some overlapping between two or more pathogens. Among the potential targets, several have been previously reported in the literature as targets for antimicrobial development, and many targets have approved drugs. Despite similarities in the metabolic network structure for closely related bacteria, we show that the method is able to selectively identify targets in pathogenic versus non-pathogenic organisms. Conclusions: This new computational system can give insights into the identification of new candidate therapeutic targets for pathogenic bacteria and discovery of new antimicrobial drugs through genome-scale metabolic network analysis and heterogeneous data integration, even for non-curated or incomplete networks.