search

kkang7 / CDSeq_R_Package

CDSeq R Package
17 stars 10 forks source link

Including known cell type proportions in CDSeq initial deconvolution #6

Closed emmawwinchester closed 3 years ago

emmawwinchester commented 3 years ago

Hi there! I am using your package and have run into an issue. We know that there are about 7 different cell types in our bulk samples (based on scRNAseq profiling), with one cell type taking up 70-80% of the sample. When we apply your deconvolution method, we are only able to pick up that one cell type, even when we increase the number of cell types to many greater than 7 (say to 10-15). Is there a way to input known cell type proportions in the initial CDSeq() step? Thank you!

kkang7 commented 3 years ago

Hi, currently, there is no explicit way of providing cell type proportion ranges in the bulk. I would recommend increasing the number of cell types even larger (for example, set cell type number to be 5:30) and see what the estimated number of cell types is and if other cell types could be detected. Make sure the dilution factor is set to be 1. CDSeq may find a couple of cell types with the same identity if it constitutes a large proportion in bulk samples, then those cell types can be merged into one major cell type in the cell type annotation procedure. Another thing you could try is to decrease hyperparameter alpha to be less than 1 (say 0.1 or something). alpha less than 1 is saying that the proportions of cell types are likely to be dominated by a couple of cell types. You can play with these options and see if the minority cell types could be found.

On Wed, Mar 17, 2021 at 11:43 AM emmawwinchester @.***> wrote:

Hi there! I am using your package and have run into an issue. We know that there are about 7 different cell types in our bulk samples (based on scRNAseq profiling), with one cell type taking up 70-80% of the sample. When we apply your deconvolution method, we are only able to pick up that one cell type, even when we increase the number of cell types to many greater than 7 (say to 10-15). Is there a way to input known cell type proportions in the initial CDSeq() step? Thank you!

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/kkang7/CDSeq_R_Package/issues/6, or unsubscribe https://github.com/notifications/unsubscribe-auth/AEDIUAMVSME2PM4YKKCBTADTEDE2RANCNFSM4ZK2ZF3A .