In silico deconvolution of small cancer cell ratios using transcriptomics gene signatures

Abstract: In this work we suggest a deconvolution method, based on a convex optimization problem, to calculate the cancer amount from heterogeneous cell type gene expression profiles generated in silico. Expression profiling is a technique for identifying global expression patterns within cellular groups, its multiple purposes may include the identification of disease biomarkers and the basic understanding of cellular processes. Given the necessity for understanding complex biological processes such as development and carcinogenesis, it is of main importance to distinguish between contributions to gene expression profiles from either regulation processes or abundance of cellular groups. Unfortunately, many biological samples contain mixtures of cell-types. This severely limits the conclusions that can be made about the specificity of gene expression in the cell-type of interest. We describe a model to estimate the proportions of cell types in a given test data set based on a gene expression profile derived from transcriptomics. Our model is based on least squares estimation and the solution of a convex optimization problem. The technical aim is to solve an undetermined system of linear equations, which must satisfy several constraints and under a particular sparsity assumption. Cell type mixtures were simulated in silico using a special procedure based on mean and standard deviations. Variable selection was performed by Analysis of Variance (ANOVA) using “cell type” as main factor and genes were ranked by F-statistics. We tested our model in breast and liver tissues, employing four cell types (three normal and one cancerous). We also performed a bootstrap procedure to test the robustness of our method concluding that our method is stable and accurate enough to estimate cancer portions of at least 10%.