A feasible roadmap for unsupervised deconvolution of two-source mixed gene expressions

Tissue heterogeneity is a major confounding factor in studying individual populations that cannot be resolved directly by global profiling (Hoffman, et al., 2004). Experimental solutions to mitigate tissue heterogeneity are expensive, time consuming, inapplicable to existing data, and may alter the original gene expression patterns (Kuhn, et al., 2011; Shen-Orr, et al., 2010). Alternatively, various in silico methods perform basically a supervised deconvolution based on either externally-obtained constituent proportions (Shen-Orr, et al., 2010; Stuart, et al., 2004) or previously-acquired cell-specific signatures (Kuhn, et al., 2011; Lu, et al., 2003). In the earlier issues of this journal, a few articles have reported semi-supervised methods that were specifically focused on dissecting two-source mixed gene expressions. Gosink et al. used (known) expression data from a single cell type to determine the proportion (and subsequently expression profile) of each cell type in a heterogeneous sample (Gosink, et al., 2007).