Clarification in input samples

[Dustin Sokolowski]

Hello!

Thank you for the exciting tool and paper. I was interested in looking into applying Bayes-Prism to a non-cancer database with a small sample size compared to TCGA.

Specifically, I have 50 mice with 5 timepoints and 2 conditions at each timepoint. My paired scRNA-seq data is an N=2, one sample from each condition in the middle timepoint. I was hoping to look at cell-type proportions and even differences in cell-type specific expression between conditions across timepoints. I was wondering if you've tested Bayes-Prism on sample sizes of this size and of non-tumour tissue? If you have, are there any potential roadblocks to consider?

Best,
Dustin

[Tinyi Chu]

Hi Dustin,

I am not sure if my previous email reply passed through. In case if it didn't, I am pasting my reply here.

"Hi Dustin,

Thank you for your interest in our work. Sample size of the bulk RNA-seq generally should not be an issue, as in the first (initial) Gibbs sampling step, each bulk is treated independently. Only the updated sampling step might be affected. We benchmarked the human peripheral whole blood sample with N=12 (see figure 1e and f of our paper), and found both the initial and updated gibbs accurate. That being said, we are updating our method to make the updated sampling potentially more robust to rare cell types and small numbers of bulk samples (in ~one week). You are also welcome to try the updated package.

A few suggestions are as follows.

1. As for the recommended setup, if the gene expression is expected to change across time points, you may consider sub-clustering each cell type in your scRNA-seq data, and label them as cell states. Hopefully the scRNA-seq collected at the midpoint can capture the heterogeneity of transcription from early and late time points. Doing this way may make the inferred posterior more accurate.

2. I would recommend starting by deconvolving samples from each condition using the scRNA-seq from the same condition.

Let me know if there are any questions.

Best,

Tinyi"

[Dustin Sokolowski]

Hey Tinyi! I must have missed the original response. Thanks so much for following up, I really appreciate it. I'll certainly try the new package using your advice and see how it goes. Thanks again, Dustin

…

________________________________ From: Tinyi Chu ***@***.***> Sent: Tuesday, May 10, 2022 7:00 PM To: Danko-Lab/TED ***@***.***> Cc: Dustin Sokolowski ***@***.***>; Author ***@***.***> Subject: Re: [Danko-Lab/TED] Clarification in input samples (Issue #19) Hi Dustin, I am not sure if my previous email reply passed through. In case if it didn't, I am pasting my reply here. "Hi Dustin, Thank you for your interest in our work. Sample size of the bulk RNA-seq generally should not be an issue, as Hi Dustin, I am not sure if my previous email reply passed through. In case if it didn't, I am pasting my reply here. "Hi Dustin, Thank you for your interest in our work. Sample size of the bulk RNA-seq generally should not be an issue, as in the first (initial) Gibbs sampling step, each bulk is treated independently. Only the updated sampling step might be affected. We benchmarked the human peripheral whole blood sample with N=12 (see figure 1e and f of our paper), and found both the initial and updated gibbs accurate. That being said, we are updating our method to make the updated sampling potentially more robust to rare cell types and small numbers of bulk samples (in ~one week). You are also welcome to try the updated package. A few suggestions are as follows. 1. As for the recommended setup, if the gene expression is expected to change across time points, you may consider sub-clustering each cell type in your scRNA-seq data, and label them as cell states. Hopefully the scRNA-seq collected at the midpoint can capture the heterogeneity of transcription from early and late time points. Doing this way may make the inferred posterior more accurate. 2. I would recommend starting by deconvolving samples from each condition using the scRNA-seq from the same condition. Let me know if there are any questions. Best, Tinyi" — Reply to this email directly, view it on GitHub<https://urldefense.com/v3/__https://github.com/Danko-Lab/TED/issues/19*issuecomment-1122993028__;Iw!!D0zGoin7BXfl!4JIQSDy0hJv6GCikjJNlxv9tbd2PIXG64EBfKUaPOVEtc0bWzT9R_MqxmbGQ5BECeVP5KxYw9efbgsvX6st4FTuGD5EzNbGZeYM$>, or unsubscribe<https://urldefense.com/v3/__https://github.com/notifications/unsubscribe-auth/ANONDRUHYYIPZQONS2DOF2DVJLTBNANCNFSM5UMXPRVA__;!!D0zGoin7BXfl!4JIQSDy0hJv6GCikjJNlxv9tbd2PIXG64EBfKUaPOVEtc0bWzT9R_MqxmbGQ5BECeVP5KxYw9efbgsvX6st4FTuGD5Ez_0pj9_8$>. You are receiving this because you authored the thread.Message ID: ***@***.***>

________________________________ This e-mail may contain confidential, personal and/or health information(information which may be subject to legal restrictions on use, retention and/or disclosure) for the sole use of the intended recipient. Any review or distribution by anyone other than the person for whom it was originally intended is strictly prohibited. If you have received this e-mail in error, please contact the sender and delete all copies.

[Tinyi Chu]

Hi Dustin, Thank you for your interest in our work. Sample size of the bulk RNA-seq generally should not be an issue, as in the first (initial) Gibbs sampling step, each bulk is treated independently. Only the updated sampling step might be affected. We benchmarked the human peripheral whole blood sample with N=12 (see figure 1e and f of our paper), and found both the initial and updated gibbs accurate. That being said, we are updating our method to make the updated sampling potentially more robust to rare cell types and small numbers of bulk samples (in ~one week). You are also welcome to try the updated package. A few suggestions are as follows. 1) As for the recommended setup, if the gene expression is expected to change across time points, you may consider sub-clustering each cell type in your scRNA-seq data, and label them as cell states. Hopefully the scRNA-seq collected at the midpoint can capture the heterogeneity of transcription from early and late time points. Doing this way may make the inferred posterior more accurate. 2) I would recommend starting by deconvolving samples from each condition using the scRNA-seq from the same condition. Let me know if there are any questions. Best, Tinyi

…

On Tue, Apr 26, 2022 at 2:38 PM Dustin Sokolowski ***@***.***> wrote: Hello! Thank you for the exciting tool and paper. I was interested in looking into applying Bayes-Prism to a non-cancer database with a small sample size compared to TCGA. Specifically, I have 50 mice with 5 timepoints and 2 conditions at each timepoint. My paired scRNA-seq data is an N=2, one sample from each condition in the middle timepoint. I was hoping to look at cell-type proportions and even differences in cell-type specific expression between conditions across timepoints. I was wondering if you've tested Bayes-Prism on sample sizes of this size and of non-tumour tissue? If you have, are there any potential roadblocks to consider? Best, Dustin — Reply to this email directly, view it on GitHub <#19>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AB4NHS4VGL5L4JQKY7VMDLTVHAZ2JANCNFSM5UMXPRVA> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>