Systems Pharmacology and Translational Therapeutics

抄録

Conventional approaches to drug discovery and development have relied on identification of a molecular target, selection and refinement of a drug candidate and ultimately establishing its safety and efficacy in randomized trials. However, detection of such large average effects will be superseded by the need for information relevant to drug response at the individual level. Correspondingly, it has been appreciated that many drugs perturb wider biological networks than their canonical targets and that this may contribute to variability of drug response. The importance of parsing and predicting such variability is exemplified by the case of CAR-T cells in leukemia, where response can vary from apparent “cure” to death from cytokine release syndrome. Aside from cancer, most drugs are used to treat syndromes, such as pain. We have begun to parse sources of variability in the response to nonsteroidal anti-inflammatory drugs (NSAIDs) given the commonality of their use and the risk of serious cardiovascular or gastrointestinal adverse effects in perhaps 1-2% of people exposed. To realize the goal of a more precise medicine we need to parse endophenotypes by integrating diverse consequences of drug induced network perturbation. Ultimately, we can apply deep learning and artificial intelligence to such data to refine the development of algorithms predictive of drug efficacy and safety. These can be tested prospectively in randomized trials to determine their predictive utility using surrogates, such as hypertension for cardiovascular outcomes.