Danko-Lab
Deconvolution Using BayesPrism for Tumor Subtypes
Dear BayesPrism authors,
I am delighted that you have developed such an interesting package. I am currently researching the relationships between tumor subtypes. I have annotated a large database (79 samples), including microenvironment, immune, and stromal cells. I have isolated epithelial cells and annotated tumor subgroups, resulting in six cell types. I now have two Seurat objects: one is overall, including epithelial and microenvironment, and the other is a Malignant Seurat object.
Can I ignore immune cells and directly use the malignant epithelial subtypes for deconvolution? I noticed your cell.type.state, but I don’t understand how to construct this information. How should "cell.state.labels" be built? I am quite confused about this part.
Additionally, I noticed that you identify the effect of malignant programs, exclude malignant cells, and use it to explain new factor decomposition, obtaining cell proportions of subtypes. I am very interested in this because it is precisely what I want to do. However, I don’t understand how to apply my prior knowledge in this context.
I have already annotated malignant cells and identified subgroups. I want to exclude immune cells and get the proportions of the subgroups I annotated. How can I achieve this using your method?
And I don’t understand the meaning of this code,
#Not run ebd.res.myEta <- learn.embedding(bp = bp.res, eta_prior = my.eta, # A user supplied K-by-G matrix, at a raw count (non-log) scale. cycle = 50, compute.elbo = T)
How should the eta_prior parameter be used and constructed?
Thank you very much for your help.
Best regards,
Jason
"I have already resolved the issue with cell.state.labels. Now, regarding the prior knowledge eta_prior = my.eta, how should I construct it?"
Dear BayesPrism authors,
I am delighted that you have developed such an interesting package. I am currently researching the relationships between tumor subtypes. I have annotated a large database (79 samples), including microenvironment, immune, and stromal cells. I have isolated epithelial cells and annotated tumor subgroups, resulting in six cell types. I now have two Seurat objects: one is overall, including epithelial and microenvironment, and the other is a Malignant Seurat object.
Can I ignore immune cells and directly use the malignant epithelial subtypes for deconvolution? I noticed your cell.type.state, but I don’t understand how to construct this information. How should "cell.state.labels" be built? I am quite confused about this part.
Additionally, I noticed that you identify the effect of malignant programs, exclude malignant cells, and use it to explain new factor decomposition, obtaining cell proportions of subtypes. I am very interested in this because it is precisely what I want to do. However, I don’t understand how to apply my prior knowledge in this context.
I have already annotated malignant cells and identified subgroups. I want to exclude immune cells and get the proportions of the subgroups I annotated. How can I achieve this using your method?
And I don’t understand the meaning of this code,
#Not run ebd.res.myEta <- learn.embedding(bp = bp.res, eta_prior = my.eta, # A user supplied K-by-G matrix, at a raw count (non-log) scale. cycle = 50, compute.elbo = T)How should the eta_prior parameter be used and constructed?
Thank you very much for your help.
Best regards,
Jason
Hi Jason,
You may construct a K-by-G matrix, with K being the number of subtypes. You may simply sum-up the reads of all cells from each subtype, and renormalize them to sum-to-one (for each subtype).
Best,
Tinyi
On Sat, Jun 15, 2024 at 12:29 AM xianjieshen @.***> wrote:
"I have already resolved the issue with cell.state.labels. Now, regarding the prior knowledge eta_prior = my.eta, how should I construct it?"
Dear BayesPrism authors,
I am delighted that you have developed such an interesting package. I am currently researching the relationships between tumor subtypes. I have annotated a large database (79 samples), including microenvironment, immune, and stromal cells. I have isolated epithelial cells and annotated tumor subgroups, resulting in six cell types. I now have two Seurat objects: one is overall, including epithelial and microenvironment, and the other is a Malignant Seurat object.
Can I ignore immune cells and directly use the malignant epithelial subtypes for deconvolution? I noticed your cell.type.state, but I don’t understand how to construct this information. How should "cell.state.labels" be built? I am quite confused about this part.
Additionally, I noticed that you identify the effect of malignant programs, exclude malignant cells, and use it to explain new factor decomposition, obtaining cell proportions of subtypes. I am very interested in this because it is precisely what I want to do. However, I don’t understand how to apply my prior knowledge in this context.
I have already annotated malignant cells and identified subgroups. I want to exclude immune cells and get the proportions of the subgroups I annotated. How can I achieve this using your method?
And I don’t understand the meaning of this code,
#Not run ebd.res.myEta <- learn.embedding(bp = bp.res, eta_prior = my.eta,
A user supplied K-by-G matrix, at a raw count (non-log) scale. cycle =
50, compute.elbo = T)
How should the eta_prior parameter be used and constructed?
Thank you very much for your help.
Best regards,
Jason
— Reply to this email directly, view it on GitHub https://github.com/Danko-Lab/BayesPrism/issues/88#issuecomment-2169120550, or unsubscribe https://github.com/notifications/unsubscribe-auth/AB4NHS36LEIWQI7C5OUP3D3ZHO7JRAVCNFSM6AAAAABJH3QCGCVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDCNRZGEZDANJVGA . You are receiving this because you are subscribed to this thread.Message ID: @.***>
