A custom signature matrix can be created using data from purified cell populations. While the process to generate a custom matrix from expression profiles is straightforward, the performance of a custom matrix will depend on the quality of the data used to generate it. Immunophenotyping of leukocytes is a dynamic field with new immune populations continuing to be identified. Care should be taken in determining which immune “cell types” should be included in the signature matrix and which canonical markers should be used to isolate these populations. For example, it is clear that the population of “CD4-expressing T lymphocytes” encompasses heterogeneous populations with diverse functional phenotypes including naïve, memory, Th1, Th2, Th17, T-regulatory cells and T follicular helper cells. Replicates for each purified immune cell type are required to gauge variance in the expression profile (see 5.4 for further details). The platform and methods used to generate data for the signature matrix ideally should be identical to that applied to analysis of the mixture samples. While SVR is robust to unknown cell populations, performance can be adversely affected by genes that are highly expressed in a relevant unknown cell population (e.g., in the malignant cells) but not by any immune components present in the signature matrix. A simple option implemented in CIBERSORT to limit this effect is to remove genes highly expressed in non-hematopoietic cells or tumor cells. If expression data is available from purified tumor cells for the malignancy to be studied, this can be used as a guideline to filter other confounding genes from the signature matrix.