Statistical tools such as test of means and medians are often used for identifying significantly different metabolites across different groups of individuals (e.g. control versus diseased). Inverse covariance modelling methods, which detect direct relationships between metabolites of a metabolic network, is an alternative approach that is sensitive to changes in these relationships. However, inverse covariance modelling is seldom used for analysing metabolite data due to its complexity. The aim of our study was therefore to address the technical barriers associated with inverse covariance modelling. This was achieved by investigating the Latent Differential Graphical Model (LDGM) and determining whether it is a suitable candidate for detecting true dysregulation among metabolic networks. In our research, we adapted the LDGM method to analyse metabolite data and developed a method for LDGM to sensibly handle situations in which the number of observations outnumbered the number of metabolites. We tested our approach on public datasets and assessed the viability of the LDGM method by comparing the LDGM results to the results obtained from the test of means and correlations. Our results suggest that the LDGM method can effectively identify a concise list of useful relationships within a metabolic network in cases where distinct differences exist between the groups. We therefore conclude that the LDGM method can potentially be a useful tool in metabolomics that should be further investigated.