search

Danko-Lab / TED

a fully Bayesian approach to deconvolve tumor microenvironment
60 stars 10 forks source link

Help with expected input #1

Closed kvittingseerup closed 4 years ago

kvittingseerup commented 4 years ago

Thanks for providing this tool - it looks extremely promising.

I have a couple of clarifying questions:

  1. When having data from multiple patients you suggest collapsing cell types per patient. Does that mean the input.phi matrix given to run.Ted have X reference rows where X = p * c (and p= number of patients, and c = number of celltypes)?
  2. Could you give examples run times for some of the deconvolutions you have done for the paper? Just so we have a ballpark number of what to expect.
  3. Is there are minimum number of bulk samples required to deconvolute? Can TED deconvolute fx 6 samples (a 3x2 experiment)?

Cheers Kristoffer

tinyi commented 4 years ago

Dear Kristoffer,

Thank you very much for your interest in our work. Here are the answers to your questions.

  1. In general we recommend users to collapse the tumor cells in each patient, while collapsing non-malignant cells across all patients. For example, the refGBM8 used in our paper has Patient-1-Tumor-subcluster1 ,..., Patient-8-Tumor-subclusterN (60 tumor subclusters from 8 patients), pericytes, endothelial, T cell, macrophage, oligodendrocytes (5 non-malignant cells across all patients). If there is substantial heterogeneities in the non-malignant cells, one should treat each cluster of non-malignant cell separately.

  2. See attached pdf.

  3. No minimum required. TED assumes the tumor expression in each sample is conditional independent (conditional on input.phi).

Hope this helps.

Best,

Tinyi

On Wed, Jan 29, 2020 at 10:20 AM Kristoffer Vitting-Seerup < notifications@github.com> wrote:

Thanks for providing this tool - it looks extremely promising.

I have a couple of clarifying questions:

  1. When having data from multiple patients you suggest collapsing cell types per patient. Does that mean the input.phi matrix given to run.Ted have X reference rows where X = p * c (and p= number of patients, and c = number of celltypes)?
  2. Could you give examples run times for some of the deconvolutions you have done for the paper? Just so we have a ballpark number of what to expect.
  3. Is there are minimum number of bulk samples required to deconvolute? Can TED deconvolute fx 6 samples (a 3x2 experiment)?

Cheers Kristoffer

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/Danko-Lab/TED/issues/1?email_source=notifications&email_token=AB4NHS6VLC7KAIOM5IOAZITRAGNC5A5CNFSM4KNGOY22YY3PNVWWK3TUL52HS4DFUVEXG43VMWVGG33NNVSW45C7NFSM4IJSDGZA, or unsubscribe https://github.com/notifications/unsubscribe-auth/AB4NHS4HCDIZW3S3FQN5UNLRAGNC5ANCNFSM4KNGOY2Q .

kvittingseerup commented 4 years ago

Hi Tinyi

Thanks for the quick answer!

With regards to 2: I don't think the attachment was transfered to github?

With regards to 3: Did you experiment with information sharing across bulk samples - does not seem a stretch to assume the the different cell types exist in somewhat same proportions in a cohort of bulk patients?

tinyi commented 4 years ago

Dear Kristoffer,

I have just added the runtime to the github repository. Kindly check it out.

TED does not assume similar cell type proportion across bulk samples, which is often not the case in TCGA samples due to the heterogeneity of the tumor microenvironment. In fact, the posterior of the cell type fraction is so strong, and is sufficiently driven by the mixture samples themselves.

TED does however assumes the non-malignant cells have the same expression across bulk samples. This piece of shared information is used to update the initial estimates of cell type fraction.

Hope this answers your question.

Best,

Tinyi

On Fri, Jan 31, 2020 at 3:58 AM Kristoffer Vitting-Seerup < notifications@github.com> wrote:

Hi Tinyi

Thanks for the quick answer!

With regards to 2: I don't think the attachment was transfered to github?

With regards to 3: Did you experiment with information sharing across bulk samples - does not seem a stretch to assume the the different cell types exist in somewhat same proportions in a cohort of bulk patients?

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/Danko-Lab/TED/issues/1?email_source=notifications&email_token=AB4NHS6ZL5MDIUM5FXBYNPDRAPR4DA5CNFSM4KNGOY22YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEKN7DFQ#issuecomment-580645270, or unsubscribe https://github.com/notifications/unsubscribe-auth/AB4NHS2OR2MMAZ23YXCY2BTRAPR4DANCNFSM4KNGOY2Q .

kvittingseerup commented 4 years ago

Beautiful! That is a very nice idea of sharing only the normal cells.

Impressive runtimes!

Thanks!

tinyi commented 4 years ago

Dear Kristoffer,

A vignette has now been included in the github. Also, I have updated the run.Ted function so that it will now automatically collapse, align the genes on the common subset between reference and bulk, and normalized the scRNA-seq reference.

Best,

Tinyi

On Mon, Feb 3, 2020 at 4:52 AM Kristoffer Vitting-Seerup < notifications@github.com> wrote:

Closed #1 https://github.com/Danko-Lab/TED/issues/1.

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/Danko-Lab/TED/issues/1?email_source=notifications&email_token=AB4NHS3GJQIFX62BXSPRVV3RA7SMLA5CNFSM4KNGOY22YY3PNVWWK3TUL52HS4DFWZEXG43VMVCXMZLOORHG65DJMZUWGYLUNFXW5KTDN5WW2ZLOORPWSZGOWLTPTMA#event-3001481648, or unsubscribe https://github.com/notifications/unsubscribe-auth/AB4NHS63VVSGHJUD2ZVG3Z3RA7SMLANCNFSM4KNGOY2Q .

kvittingseerup commented 4 years ago

Thanks! :D