No DARs identified?

[yeswzc]

Hi, Do anyone know why I identified no DARs under the threshold of FDR<0.05 using snapATAC ?

I analyzed my human 10x genomics scATAC-seq data according to '10X Adult Mouse Brain' tutorial and tried to find DARs by:

DARs = findDAR(
        obj=x.sp,
        input.mat="pmat",
        cluster.pos=cluster_i,
        cluster.neg=NULL,
        cluster.neg.method = DAR.method, #knn or random
        bcv= 0.4,
        test.method="exactTest",
        seed.use=10
    );

However, no significant DARs were identified for each cluster. Do anyone have the same problem?

Thanks!

[r3fang]

Hello,

How many cells do you have for cluster_i?

[yeswzc]

The number of cells varied across clusters, most of them have hundreds of cells.

> sort(table(x.sp@cluster))

  10   11    2   12   13    6    1    4    3    9    7    8    5
  15  139  214  227  405  407  447  788  835  846  891 1080 1470

Thank you!

[qingnanl]

I do see the same thing for some clusters with my dataset. I tuned the parameters and found that lowering the bcv made some regions appeared as significant ones. I am wondering whether this tuning is valid, or is the dispersion itself is actually lower for ATAC-seq compared with RNA-seq. Thanks.

[yeswzc]

@qingnanl , thank you! I am also not sure if it is ok to tune bcv. I followed the vignette of 10X Adult Mouse Brain, in which the author indicated bvc = 0.4 for human, 0.1 for mouse. Hope @r3fang could help.

[suragnair]

These bcv values are for RNA-seq. There's no reason they should translate to ATAC-seq as well. The 0.4 and 0.1 values come from Section 2.11 here.