Differential expression analysis of binary appearance patterns

Abstract

Differential expression analysis of biomarker measurements in two conditions is a basic step in understanding the relevance of the selected molecule to the difference between these conditions. In cases, when the biomarker of interest can disappear completely in some of the tested samples, it can be beneficial to replace the quantitative measurements with a simple binary appearance indicator. Standard differential expression approaches, working on quantitative measurements and addressing non-existing molecules as measured with zero intensity, are designed to evaluate the change between low- and high-expression molecules, and therefore not suited well for the evaluation between existing and non-existing molecules. Here we propose two approaches to compare such binary appearance patterns in two independent groups: (i) proportion-based test and (ii) hypergeometric-distribution based test. The first leverages the fact that population proportion behaves according to the normal distribution, while the second reduces the comparison problem to the test of significance of the intersection size between two groups. We demonstrate the value of the suggested binary differential expression analysis methods by investigating (i) proteomic profiles obtained from samples extracted with ebiopsy technique and (ii) transcriptomic profiles obtained from samples extracted with standard biopsy from basal cell carcinoma and squamous cell carcinoma lesions and discuss the results with respect to raw-measurement data based differential expression test of the raw measured intensities.