It proves so far difficult to predict the metabolome, even when genome, transcriptome or proteome of a cell are known. In order to globally map enzyme-metabolite relationships, we systematically quantified enzyme expression and metabolite concentrations in Saccharomyces cerevisiae kinase knock-out strains. Enzymes expression changes did account for a major fraction of all differentially expressed proteins, and were non-redundant, implying that kinases act generally yet specifically in metabolic regulation. Differential enzyme expression was found to affect metabolite concentrations through the redistribution of flux control, resulting in a many-to-many relationship between enzyme abundance and the metabolome. Machine learning successfully mapped these relationships, allowing the precise prediction of metabolite concentrations, as well as identifying regulatory genes. Our study reveals that hierarchical metabolic regulation acts predominantly through adjustment of broad enzyme expression patterns rather than over rate-limiting enzymes, and may account for more than half of metabolic regulation.