The Gene Expression Deconvolution Interactive Tool (GEDIT): Accurate Cell Type Quantification from Gene Expression Data

Abstract

The cell type composition of heterogeneous tissue samples can be a critical variable in both clinical and laboratory settings. However, current experimental methods of cell type quantification (e.g. cell flow cytometry) are costly, time consuming and can introduce bias. Computational approaches that infer cell type abundance from expression data offer an alternate solution. While these methods have gained popularity, most are limited to predicting hematopoietic cell types and do not produce accurate predictions for stromal cell types. Many are also limited to particular platforms, whether RNA-Seq or specific microarray models. To overcome these limitations, we present the Gene Expression Deconvolution Interactive Tool, or GEDIT. Using simulated and experimental data, we demonstrate that GEDIT produces accurate results for both stromal and hematopoietic cell types. Moreover, GEDIT is capable of producing inputs using RNA-Seq data, microarray data, or a combination of the two. Finally, we provide reference data from 7 sources spanning a wide variety of stromal and hematopoietic types. GEDIT also accepts user submitted reference data, thus allowing deconvolution of any cell type, provided that accurate reference data is available.