These studies likewise used cell populations as a source of phenotypic data; by contrast, we emphasize the need to ultimately consider single cells. These studies also stop short of using quantitative approaches to suggest a framework in which cancer drugs themselves are classifiers. In contrast to these lines of research, we highlight a process by which potential targets for drug optimization can be identified by defining discriminability as the drug’s objective. We use this property as a rationale for drug combinations that should optimize drug efficacy. We also demonstrate that one can design optimal drug classifiers using single-cell data that reflects normal and tumor cell heterogeneity. Conceptualizing drugs as classifiers is not only meaningful for cancer treatments. Any drug that should produce a binary outcome could be modeled using the same framework. This framework should generalize to drugs, such as those responsible for asthma attacks. Because differences between healthy and cancerous tissues may be subtle, we sought to explore differences between the most similar groups of cells we could find in the data set. The original study used principal component analysis to identify sub-populations within the healthy, primary tumor, and xenograft tissues. We chose one particular sub-population that existed in both healthy and primary tumor tissues: stem-like cells. We then tested whether we could use a GLM to reliably identify which cells belonged to the healthy tissue and which belonged to tumor tissue, and how many markers were needed to do so. We used another dataset to ask if actual cancer drugs act as classifiers and to determine how to optimize cancer treatment. They measured the expression of approximately 19,000 genes in breast cancer tissues using microarrays, and the chemotherapeutic responses of those tissues. The breast cancer tissue came from a panel of 45 breast cancer lines. After gene expression of each cell line was measured, each line was treated with one of 74 drugs. They defined the sensitivity of a cell line to a given drug as the concentration of drug at which 50% of cell growth was inhibited. We thus have a dataset where we know the markers and we know how strongly the cells responded to a variety of drugs. Metastasis is a multistep process including invasion of the surrounding tissue, intravasation, survival in the SB431542 citations bloodstream, extravasation and colonization of distant sites. For the first steps in this process, cancer cells frequently switch from a sessile, epithelial phenotype towards a motile, mesenchymal phenotype, a process called epithelial-to-mesenchymal transition. In cancer, aberrant activation of this latent embryonic program contributes to progression to metastatic disease and therapeutic resistance, enabling cancer cells to become invasive, disseminate, resist apoptosis, stimulate angiogenesis and acquire stem/progenitor cell properties.