|
Abstract:
|
To improve immunotherapy response, one crucial step is to study the immune/stromal cell composition in the tumor microenvironment. Deconvolution analyses based on transcriptomic data have been utilized to estimate cell composition from bulk data. However, there is no proper deconvolution algorithm suitable for proteomics data, hindering the relevant investigation in the fast growing proteogenomic research. To bridge this gap, we propose BayesDeBulk, a flexible deconvolution method which can be applied to estimate the immune/stromal composition from gene expression as well as proteomic profiling. A challenge when performing deconvolution analysis based on proteomic data is that the abundance levels across different proteomic datasets are in general not comparable. This makes the reference-based deconvolution methods not suitable. To overcome this problem, we propose a novel reference-free Bayesian model. The idea is to leverage the information of known cell-type-specific markers without using their absolute abundance levels from purified cells. Improved performances of BayesDeBulk over Cibersort and xCell are shown on different synthetic and real data examples.
|
