Using gene set signature to guide hypothesis-driven experiments

Complex diseases, such as cancer, are inherently multi-genetic mutations. Since it may be caused by a combinatorial effect of many mutations, the individual effect of each mutation may be too small to discover. In addition, the progresses of tumorigenesis not only include the mutations that facilitate tumorigenesis (called tumor drivers), but also include those accumulated during the growth of the tumor (known as tumor passengers) (1). Therefore, high degree of morphologic and clinical diversities exists among various types of cancers; intrinsic heterogeneity of cancer is very common among patients due to different genetic and environmental perturbations. However, traditional approaches of clinical disease classification are mainly based on pathological analysis of patients and existing knowledge of diseases. The current pathologic classification and ability to predict postsurgical prognosis are quite inadequate. Hence, although the existence of marked heterogeneity is well appreciated, virtually all cancers currently are treated similarly.