CellCODE: a robust latent variable approach to differential expression analysis for heterogeneous cell populations

Abstract

In this study, we consider a clinically relevant situation where neither accurate proportion estimates nor pure cell expression is of direct interest, but where we are rather interested in detecting and interpreting relevant differential expression in mixture samples. We develop a method, Cell-type COmputational Differential Estimation (CellCODE), that addresses the specific statistical question directly, without requiring a physical model for mixture components. Our approach is based on latent variable analysis and is computationally transparent; it requires no additional experimental data, yet outperforms existing methods that use independent proportion measurements. CellCODE has few parameters that are robust and easy to interpret. The method can be used to track changes in proportion, improve power to detect differential expression and assign the differentially expressed genes to the correct cell type.

Enumerateblood – an R package to estimate the cellular composition of whole blood from Affymetrix Gene ST gene expression profiles

“…More recently, reference-free approaches to quantifying the composition of mixed tissue samples from gene expression [13] or DNA methylation [14] profiles have been proposed. Such approaches may offer a solution to the platform mapping issues we describe.…”

“…Reference-free approaches rely on the availability of marker genes for the cell types to be quantified. Many strategies have been described for identifying such marker genes [5, 12, 13, 28], but, so far, all have leveraged existing reference datasets: collections of gene expression profiles derived from cells isolated from the mixed tissue to be quantified. In order to determine whether marker gene selection exhibits platform-bias, we compared CellCODE SPVs derived using either marker genes identified from the IRIS reference dataset (U133 platform; mapped to Gene ST platform identifiers) or the features in our model.…”

“…The reference-free approach described by Chikina et al [13], does not require a basis matrix, relying instead on a set of putative marker genes. These are used to guide the decomposition of the dataset’s covariance structure into separate variance components, using singular value decomposition.…”

“…More recently, so-called reference-free approaches have been proposed to address this issue [13–15]. These approaches still require the identification of suitable marker genes for the cell types to be quantified, however, and this selection is of paramount importance to achieve optimal performance.…”

“…These approaches still require the identification of suitable marker genes for the cell types to be quantified, however, and this selection is of paramount importance to achieve optimal performance. The general strategy for marker selection is to identify genes whose expression in one cell type differs from that observed in all other cell types being considered [13], a process that itself relies on reference datasets. In fact, all approaches discussed thus far leverage one of a handful of publicly available reference datasets to derive a basis matrix or identify suitable marker genes [11, 16, 17].…”