Combination chemotherapies have been a mainstay in the treatment of disseminated malignancies for almost 60 y, yet even successful regimens fail to cure many patients. Although their singledrug components are well studied, the mechanisms by which drugs work together in clinical combination regimens are poorly understood. Here, we combine RNAi-based functional signatures with complementary informatics tools to examine drug combinations. This approach seeks to bring to combination therapy what the knowledge of biochemical targets has brought to single-drug therapy and creates a statistical and experimental definition of "combination drug mechanisms of action." We show that certain synergistic drug combinations may act as a more potent version of a single drug. Conversely, unlike these highly synergistic combinations, most drugs average extant single-drug variations in therapeutic response. When combined to form multidrug regimens, averaging combinations form averaging regimens that homogenize genetic variation in mouse models of cancer and in clinical genomics datasets. We suggest surprisingly simple and predictable combination mechanisms of action that are independent of biochemical mechanism and have implications for biomarker discovery as well as for the development of regimens with defined genetic dependencies.chemotherapy | lymphoma | RNAi signature | systems biology C urrent rationales for the design of combination chemotherapy regimens were developed in the 1940s and 1950s (1-5) and, remarkably, were concurrent with the identification of DNA as the genetic material. The development of these regimens in the absence of any knowledge of cancer genomics was a remarkable achievement. However, although our knowledge of the genetic drivers of cancer and the mechanisms of drug action has increased dramatically over the past 30 y, this information has been difficult to adapt to clinical practice. As such, even very successful combination regimens often fail to cure many patients (6, 7). We hypothesize that part of this failure is due to the absence of mechanistic information about how drugs in regimens interact to promote combination effects (we term these effects "combination mechanisms of action"). We sought to address this gap by investigating specific hypotheses concerning the "mechanisms of action" of combination therapy.The classic term "drug mechanism of action" refers to the description of a specific biochemical event, which is often the activation or inhibition of an enzymatic effect. However, in recent years, "signature"-based prediction has provided a powerful new strategy for examining drug mechanism. In signature-based approaches, a series of drug-induced molecular/phenotypic measurements are made in an experimental system. Collections of measurements from many small molecules form multivariate signatures that aim to fingerprint drugs based on their relative signature similarity (8-13). In several landmark studies using Saccharomyces cerevisiae; gene expression compendia (8); and, later, barcoded loss o...