Inferring metabolite states from spatial transcriptomes using multiple graph neural
Metabolism serves as the pivotal interface connecting genotype and phenotype in various contexts, such as cancer reprogramming and immune metabolic reprogramming. Compared to the transcriptome, the development of the single-cell metabolome faces significant challenges. While various methods exist for predicting metabolite levels from transcriptome, their efficacy remains limited. We developed an efficient and adaptable algorithm known as Multiple Graph-based Flux Estimation Analysis (MGFEA). MGFEA enables rapid inference from million-level single-cell transcriptome datasets and achieves accuracy comparable to that of scFEA. Additionally, MGFEA can detect metabolite biomarkers in different cancer bulk RNA-seq datasets. As an attempt to integrate multi-omics dataset, MGFEA can further improve the accuracy of these inferences by leveraging additional metabolome.