Metabolism is an important regulator of cancer behavior. The extent to which patient-to-patient variability in tumor metabolism influences patient prognosis, however, remains unclear.
We addressed this question by developing an approach to simulate cancer metabolism that integrates RNA-seq data with genome-scale reconstructions of human metabolism into a numerically stable mathematical model of tumor metabolism in individual patients. This approach mitigates the effects of patient-to-patient RNA-seq variability by “projecting” RNA-seq data through a connected, functional network and integrates the coordinated effects of RNA-expression through the metabolic network, greatly increasing the biological interpretability of RNA-seq data.
We tested this approach by simulating metabolism in over 2,000 individual patients with diverse tumors including breast, lung, prostate, pancreatic, head and neck, and bladder cancers. We demonstrated that this approach can predict tumor aggressiveness directly from RNA-seq data. Additionally, classification of tumors based on simulated metabolism identified metabolic subtypes significantly associated with overall survival prognosis.