Entering edit mode
3.7 years ago
garciajessica0311
▴
10
Hi, I performed single-cell RNA sequencing of 26 samples. I have fold-change values for all samples, with my genes of interest (n=57). The data has been previously filtered and normalized, . I represented my data in an heatmap but I would like to show that genes are not differentially expressed between my samples. What is the best statistical methodology ? Thanks for your help, Best, Jessica
When you say 26 samples, do you mean 26 cells, or many cells from 26 samples? When you say you want to show that htey are not DE, do you mean you want a test to show each each gene is not differentially expressed, or you want to test whether they, as a group, are not differentially expressed.
I have 26 cells. Yes, I looking for a test that shows that the gene is equally expressed in each cell.
How was the data
filtered and normalized? It sounds like you are cherry picking genes you want, is that correct?The data was filtered with CPM and normalized with VOOM. From normalization step, I have a table with normalized log values for all filtered genes. Among filtered genes, I have filtered the 57 genes.
What type of single-cell sequencing was this? Chromium? Plate-based?
Where do you derive the fold change values from, i.e. what type of comparison have you done?
It's a template-switching method that enables to generate full-lenght cDNA. Sorry, it's log values
typically, one identifies differentially expressed genes by comparing the values of multiple replicates for, say, two conditions. E.g. expression values from 6 MUT samples vs. 6 WT samples. The replicates are needed because they are assumed to capture the typical biological variability that is to be expected for the expression measurements of a single gene across multiple samples and the null hypothesis typically states that there is no systematic difference between the two groups.
As the many questions and comments illustrate, it is not entirely clear to us what the outcome of your analysis is supposed to be; we'll need more details around your experiment and the underlying question of interest. Do you wish to say that the gene expression values you're capturing across your 26 cells are consistent and in the realm of biological variability? Do you have a set of control genes where you know you should be seeing differences? Why do you not have different conditions that could be compared?
To clarify this experiment : I used cell line model to perform scRNAseq. My workflow is divided in 3 parts (time-point) and I recovered cells (at single-cell level or small bulk) at each time-point : - TP1 : triplicates - TP2 : triplicates - TP3 : 20 samples (3 bulks and 17 single-cells) The end-point is to evaluate the impact of the workflow on my cells by comparing transcriptomic data at each time point, including apoptosis pathway. I have filtered 57 genes involved in apoptosis pathway. I am looking for the best statistical method to evaluate whether these genes are differentially expressed or not. We can considered that TP1 is the reference.
I'm sorry, you lost me; I cannot envision the experimental set up you're describing.
That being said, to test for differential gene expression, you can go several routes. With the few samples that you have, you could either use limma (since you already voom-transformed the data anyway) or you can follow the protocol described in the OSCA resource. Hope that helps!
Against what were those values compared to get fold changes?
Sorry, it's log values not fold-change values