Figure 2: Illustration of the unmixing approach. Tumor samples T1–T4 are assayed by aCGH, generating genome-wide copy number profiles. The aCGH profiles are interpreted as points in a space (two-dimensional in the example) and are unmixed by fitting a simplex to the point set (a 3-simplex, or triangle, in the example). The vertices of the simplex represent inferences of three cell types (1, 2, and 3) from which T1–T4 can be explained. These vertices are then projected back to the dimension of the aCGH arrays to construct virtual aCGH profiles of the inferred cell types. The outputs are these virtual aCGH profiles and the inferred fractional amount of each cell type in each tumor sample.