Re: High proportion of genes with non-independent bursting

[Jiang, Yuchao]

Hi Roy,

SCALE needs to apply to a homogeneous population of cells, where the assumption is that the same set of bursting kinetics are shared across all cells. Coordinated bursting can sometimes be used as as way to measure the possible heterogeneity within the pool of cells. For example, if there are two lineages/subgroups of cells, each having specific genes that are expressed uniquely within their respective group, these genes will stand significant in the testing of coordinated bursting. You can read more in the discussion section in our paper where we talk about the potential effect of heterogeneity on testing results.

Therefore, before applying SCALE to the dataset, one needs to make sure of the homogeneity across the cells. In your dataset, when I apply PCA on the allelic ratios (proportion of reads from alleleA), I see a clear pattern of two groups of cells. In this case, you need to apply SCALE to each of these two groups separately. I also tried tSNE, the pattern is less obvious. The code I used for your dataset is attached.

I will update our manual on data cleaning.

Cheers,

Yuchao

On May 8, 2017, at 9:43 PM, Roy Moh Lik Ang <ra...@stanford.edu> wrote:

Hi Yuchao,

 

Sorry for flooding you with so many emails. SCALE has been tremendously helpful with my work so far and I am really grateful for your involvement in this process.

 

I took a look at the P-values from the tests of independence in allele-specific bursting in the fibroblast data, and noted a substantially higher number of genes that are significant for non-independent bursting. In this case, I have not done any FDR control but simply took any genes with P-value < 0.00001 in both the mouse blastocyst data analyzed in your paper and my data (see bar chart below). Applying BH adjustment using p.adjust does little to change the overall proportion that test significant.

 

<BBFB46B740E949349BA6E0E559B06908.png>

 

I was wondering if you might know of possible explanations for this observation, of why I am observing so many significant genes with non-independent allelic bursting when the literature states most alleles should undergo independent bursting? Should I look to increase my significance threshold with respect to other parameters, such as sequencing depth or number of cells used in the measured population?

Thank you for your time. I appreciate your help.

 

Regards,

Roy