Prediction of metabolic changes that result from genetic or environmental perturbations has several important applications, including diagnosing metabolic disorders and discovering novel drug targets. A cardinal challenge in obtaining accurate predictions is the integration of transcriptional regulatory networks with the corresponding metabolic network. We propose a method called probabilistic regulation of metabolism (PROM) that achieves this synthesis and enables straightforward, automated, and quantitative integration of high-throughput data into constraint-based modeling, making it an ideal tool for constructing genome-scale regulatory-metabolic network models for less-studied organisms. PROM introduces probabilities to represent gene states and gene-transcription factor interactions. By using PROM, we constructed an integrated regulatory-metabolic network for the model organism, Escherichia coli, and demonstrated that our method based on automated inference is more accurate and comprehensive than the current state of the art, which is based on manual curation of literature. After validating the approach, we used PROM to build a genome-scale integrated metabolic-regulatory model for Mycobacterium tuberculosis, a critically important human pathogen. This study incorporated data from more than 1,300 microarrays, 2,000 transcription factor-target interactions regulating 3,300 metabolic reactions, and 1,905 KO phenotypes for E. coli and M. tuberculosis. PROM identified KO phenotypes with accuracies as high as 95%, and predicted growth rates quantitatively with correlation of 0.95. Importantly, PROM represents the successful integration of a top-down reconstructed, statistically inferred regulatory network with a bottom-up reconstructed, biochemically detailed metabolic network, bridging two important classes of systems biology models that are rarely combined quantitatively.