Discrepancies Between matched Single-cell and Bulk Transcriptome Samples

[Rui-Jing]

Hi Tinyi,

Firstly, I would like to express my appreciation for developing BayesPrism, a user-friendly software that has significantly aided my research. However, I have encountered some issues during its application in tumor sample analysis that I hope to discuss with you:

Inconsistency in Deconvolution Results: My single-cell data is derived from tumor samples. I noticed that when I increased the number of single-cell reference samples from 4 to 8, the deconvolution results became inconsistent. This was particularly evident with some types of immune cells, such as T cells, where the results varied before and after updating the single-cell samples.

Overestimation of Microenvironment Cells: In the tumor cell deconvolution results, some microenvironment cells, such as Ependymal cells, appear to be significantly overestimated. In some samples, these cells were estimated to make up 25-35% of the population, which seems implausible.

Discrepancies Between Single-cell and Bulk Transcriptome Samples: I have samples for which both single-cell and bulk transcriptomes were analyzed. When deconvolving these samples, I found considerable differences in the cell proportions estimated from the two types of data.

Lastly, I would like to know how to effectively evaluate the accuracy and reasonableness of the deconvolution results. I look forward to your guidance and thank you in advance for your response.

[tinyi]

Hi Rui-Jing, Sorry for the delay. Thank you for your feedback. I would be happy to help troubleshoot. Could you please provide additional details on how you specify the cell.type.labels and cell.state.labels? You may provide this information using the table function in R. Regarding the overestimation of Ependymal cells, one possible explanation could be their similarity to tumor cells. This can occur when tumor cells overexpress Ependymal-specific genes, but there is a lack of this cell type/state in the reference due to tumor heterogeneity. To mitigate this issue, you might consider running deconvolutions on marker genes. Best, Tinyi

On Fri, May 24, 2024 at 2:16 AM Rui-Jing ***@***.***> wrote: Hi Tinyi, Firstly, I would like to express my appreciation for developing

BayesPrism, a user-friendly software that has significantly aided my research. However, I have encountered some issues during its application in tumor sample analysis that I hope to discuss with you:

Inconsistency in Deconvolution Results: My single-cell data is derived

from tumor samples. I noticed that when I increased the number of single-cell reference samples from 4 to 8, the deconvolution results became inconsistent. This was particularly evident with some types of immune cells, such as T cells, where the results varied before and after updating the single-cell samples.

Overestimation of Microenvironment Cells: In the tumor cell deconvolution

results, some microenvironment cells, such as Ependymal cells, appear to be significantly overestimated. In some samples, these cells were estimated to make up 25-35% of the population, which seems implausible.

Discrepancies Between Single-cell and Bulk Transcriptome Samples: I have

samples for which both single-cell and bulk transcriptomes were analyzed. When deconvolving these samples, I found considerable differences in the cell proportions estimated from the two types of data.

Lastly, I would like to know how to effectively evaluate the accuracy and

reasonableness of the deconvolution results. I look forward to your guidance and thank you in advance for your response.

…

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you are subscribed to this thread. Message ID: ***@***.***>