search

Danko-Lab / BayesPrism

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

Estimating the proportion of 'unknown' cells. #47

Open alnnch opened 1 year ago

alnnch commented 1 year ago

Hi,

I was wondering if there is a way to estimate the proportion of cells that aren't present in the reference cells? I've deconvoluted bulk RNA-seq data for a tumour with a reference atlas from the same healthy tissue. There is no single cell data on this type of cancer, so I was wondering if I could predict gene expression of cancer cells from BayesPrism output? For example, EPIC gives a proportion of 'otherCells' in its output, so I was wondering if I could get something like this from BayesPrism.

Many thanks, Alina

tinyi commented 1 year ago

Hi Alina,

Thank you for your interest in our method.

The deconvolution module of BayesPrism does not infer any cells unobserved from the reference. For it to work, it is important to get the reference that best represents the cell type present in the mixture. We do not recommend deconvolving bulk tumor samples using a reference atlas from healthy tissue.

However, the embedding learning module infers the tumor cell using a linear combination of factors (while fixing the tumor fraction). That being said, it still depends on an accurate estimation of other non-tumor cell type fractions estimated from the deconvolution module.

Best,

Tinyi

On Thu, Jul 13, 2023 at 8:02 AM alnnch @.***> wrote:

Hi,

I was wondering if there is a way to estimate the proportion of cells that aren't present in the reference cells? I've deconvoluted bulk RNA-seq data for a tumour with a reference atlas from the same healthy tissue. There is no single cell data on this type of cancer, so I was wondering if I could predict gene expression of cancer cells from BayesPrism output? For example, EPIC gives a proportion of 'otherCells' in its output, so I was wondering if I could get something like this from BayesPrism.

Many thanks, Alina

— Reply to this email directly, view it on GitHub https://github.com/Danko-Lab/BayesPrism/issues/47, or unsubscribe https://github.com/notifications/unsubscribe-auth/AB4NHSYYLYNRALFUC6WGFMLXP7PTLANCNFSM6AAAAAA2I2X7VE . You are receiving this because you are subscribed to this thread.Message ID: @.***>