search

Danko-Lab / BayesPrism

A Fully Bayesian Inference of Tumor Microenvironment composition and gene expression
153 stars 46 forks source link

Using sorted bulk data as reference #44

Open AlmogAngel opened 1 year ago

AlmogAngel commented 1 year ago

Dear developers,

First off, I'd like to express my appreciation for the development of this method.

I'm interested in using TPM normalized bulk references of sorted cell types. This includes both RNA-seq and micro-array based references. According to your tutorial, it seems this can be accomplished by setting input.type = "GEP".

However, I am not sure what would be the correct way to generate markers for each type. The get.exp.stat and select.marker functions are designed for scRNA-seq data.

A potential workaround I'm considering manually performing DE analysis in a pairwise manner for each cell type, and then selecting the most discriminating genes as markers. However, I would like to verify if this approach is okay or if there's a better method available.

Any advice or guidance you can offer would be greatly appreciated.

Almog

tinyi commented 1 year ago

Hi Almog,

Thank you for your interest in our method.

Yes. You can perform DE analysis by comparing each cell type vs every other cell type, and then select the genes of higher expression level in the cell type than all other cell types, which corresponds to the "all marker" in the find.marker function provided by scran. In our hands, "all marker" works better than "any marker" (higher expressed than at least one other cell type).

We have not tested if microarray data can be used as reference. We noted one recent study that applied BayesPrism to deconovovle the microarray bulk (not the microarray reference), which seems to perform well. You may give it a try and see if the results make sense.

Best,

Tinyi

On Mon, Jun 26, 2023 at 7:21 AM Almog Angel @.***> wrote:

Dear developers,

First off, I'd like to express my appreciation for the development of this method.

I'm interested in using TPM normalized bulk references of sorted cell types. This includes both RNA-seq and micro-array based references. According to your tutorial, it seems this can be accomplished by setting input.type = "GEP".

However, I am not sure what would be the correct way to generate markers for each type. The get.exp.stat and select.marker functions are designed for scRNA-seq data.

A potential workaround I'm considering manually performing DE analysis in a pairwise manner for each cell type, and then selecting the most discriminating genes as markers. However, I would like to verify if this approach is okay or if there's a better method available.

Any advice or guidance you can offer would be greatly appreciated.

Almog

— Reply to this email directly, view it on GitHub https://github.com/Danko-Lab/BayesPrism/issues/44, or unsubscribe https://github.com/notifications/unsubscribe-auth/AB4NHSZBCSCDPHSQJZK3OR3XNFWDNANCNFSM6AAAAAAZUANL7E . You are receiving this because you are subscribed to this thread.Message ID: @.***>