Question regarding Cell Type-Specific Gene Expression Matrix (Z) from BayesPrism

[max5iu]

T celll-type.csv

T celll-state.csv

Dear BayesPrism Developers

First and foremost, we would like to express our sincere gratitude for developing BayesPrism. It's an excellent tool that has been incredibly helpful for our research.

We are using BayesPrism to analyze data from human liver cancer samples. While examining the results, specifically the cell type-specific gene expression count matrix (denoted as Z), we encountered some observations that we would like to inquire about:

Low expression of T-cell markers: Within the cell population annotated as T cells, key T-cell marker genes that we expected to be highly expressed (e.g., CD3D, CD3E) show very low expression levels, near zero, in matrix Z. We have confirmed that these genes are clearly expressed in the T-cell populations within the input reference single-cell expression profile used for deconvolution.
High expression of tumor markers in non-tumor cells: Concurrently, we observed that known tumor marker genes (e.g., EPCAM, KRT8, KRT18, KRT19) appear to have relatively high expression levels in T cells and potentially other non-tumor cell types within matrix Z.
We noted that these observations seem consistent regardless of whether we examine results based on 'cell state' or 'cell type' annotations provided by the tool.

Based on these observations, we have a couple of questions:

Is the cell type-specific gene expression matrix Z primarily designed to focus on the state of tumor cells or the interactions between tumor and other cell types? Could this focus potentially lead to a representation where the intrinsic expression signatures of certain cell types (like T cells) are downplayed or altered compared to their reference profiles?
If our primary research interest lies in deeply analyzing the gene expression profiles of specific non-tumor cell types (such as T cells, B cells, macrophages) within the tumor microenvironment, is directly using matrix Z the most appropriate approach? Alternatively, are there other outputs or perhaps specific analysis modules within BayesPrism that you would recommend for this purpose?
Thank you very much for your time and assistance with these questions. We appreciate any insights you can provide.

Sincerely,
Maxmillion

[max5iu]

Z.Bcell=get.exp(bp=bp.res,
state.or.type = "state",
cell.name = " T cell")
head(Z.Bcell)
write.csv(Z.Bcell, ' T celll-2.csv')

To help us understand these observations, could you please clarify whether the specific marker genes we mentioned (CD3D, CD3E for T cells) were included in the gene set utilized by BayesPrism during the deconvolution process? We want to ensure these genes were indeed part of the analysis leading to matrix Z, as their potential exclusion could explain the observed expression patterns.