tinyi
commented
1 month ago Dear user,
Thank you for your question.
When clustering using deconvolved cell type-specific gene expression, it is typically expected to observe lower variance than the un-deconvolved data, and hence fewer clusters. This is because the deconvolved cell type-specific gene expression has removed the variance due to the changes in the cell type abundance, and focuses on variance in the expression of a single cell type.
One suggestion for performing downstream analysis using deconvolved cell type-specific gene expression is to subset on samples of high abundance of the cell type, as this will yield more confident inference of cell type-specific gene expression profile. You may refer to Fig 1h in the BayesPrism paper for the details of such effects.
Best,
Tinyi
On Thu, Jul 25, 2024 at 9:53 AM JaneeeeeeW @.***> wrote:
Hi there, I have used BayesPrism to deconvolute my datasets utilizing reference data from a public database. My datasets consist entirely of human bone marrow mesenchymal stem cells and are not cancer samples. After obtaining the deconvolution results from BayesPrism, I reorganized the data and imported it into a Seurat object. I aimed to generate a UMAP plot resembling the reference scRNA-seq data. However, I observed that the deconvoluted results displayed fewer clusters compared to the reference data. I am seeking advice on whether it is feasible to achieve the same number of clusters as observed in the reference scRNA-seq data. Additionally, I would appreciate insights on whether aiming for the same cluster count makes sense in this context. Thank you very much for your assistance.
— Reply to this email directly, view it on GitHub https://github.com/Danko-Lab/BayesPrism/issues/94, or unsubscribe https://github.com/notifications/unsubscribe-auth/AB4NHS66ETZSWZ2IMRSJGDDZOD7O7AVCNFSM6AAAAABLOTRZQOVHI2DSMVQWIX3LMV43ASLTON2WKOZSGQZTAMBTGYZDIMA . You are receiving this because you are subscribed to this thread.Message ID: @.***>
Hi there, I have used BayesPrism to deconvolute my datasets utilizing reference data from a public database. My datasets consist entirely of human bone marrow mesenchymal stem cells and are not cancer samples. After obtaining the deconvolution results from BayesPrism, I reorganized the data and imported it into a Seurat object. I aimed to generate a UMAP plot resembling the reference scRNA-seq data. However, I observed that the deconvoluted results displayed fewer clusters compared to the reference data. I am seeking advice on whether it is feasible to achieve the same number of clusters as observed in the reference scRNA-seq data. Additionally, I would appreciate insights on whether aiming for the same cluster count makes sense in this context. Thank you very much for your assistance.