Description of the Method
The method Estimating the Proportion of Immune and Cancer cells (EPIC) was published in eLife and the original method is described here. We have used the method implementation from immunedeconv, which is a systematic benchmarking study of widely used supervised cell type deconvolution algorithms. We have made the method available in the Docker image1.
EPIC is a supervised method which uses curated gene expression signature matrix \((S)\) of 5 immune cell types (B cell, CD4+/CD8+ T cell, Macrophage, NK cell) and 2 stromal cell types (Cancer associated fibroblast and Endothelial cells) to estimate the relative abundances \((P)\) of these cell types in given transcriptomic profiles \((E)\).
Essentially, the following equation is solved using constrained least square regression where \(g\) is genes, \(s\) is samples and \(c\) is cell types
\[E_{gs} = S_{gc}.P_{cs}\]
Results
The program returns an estimated proportion matrix and the corresponding reference cell-type specific profiles matrix.
The cell type estimates are comparable across samples and also within a sample for different cell types.
Descriptors
Omics_type = trancriptome
Cancer_type = all
Method_type = supervised
Cell_type = immune