Correct approach to identify differential binding in samples (with no duplicates) from two subtypes of a cancer

Question

[Researcher]

Hi all,

I want to quantitatively compare ChIP-seq data between different subtypes of a cancer.

My experimental design is as follows: I have multiple samples enriched for 3 different histone marks from 2 subtypes of a cancer and there are no replicates for them. The sample structure somewhat looks like this:

Subtype 1                                          

Sample_A_K4me3, Sample_A_K27me3, Sample_A_K27ac, Sample_1_input, 
Sample_B_K4me3, Sample_B_K27me3, Sample_B_K27ac, Sample_2_input, 
Sample_C_K4me3, Sample_C_K27me3, Sample_C_K27ac and Sample_3_input

Subtype 2

Sample_1_K4me3, Sample_1_K27me3, Sample_1_K27ac, Sample_1_input, 
Sample_2_K4me3, Sample_2_K27me3, Sample_2_K27ac, Sample_2_input, 
Sample_3_K4me3, Sample_3_K27me3, Sample_3_K27ac and Sample_3_input

Since, I am interested to call differential binding sites between the subtypes of a cancer, Can I use these samples (like samples A,B and C) which are enriched for same histone mark (say K4me3) from same subtype as a surrogate to replicates and compare it with the similar set (1,2 and 3) from other subtype?

If this is ok which tool is recommended to use in order to compare these unpaired samples and call differential binding sites between both the subtypes.

Thanks in advance.

[ATpoint]

Are for example Sample_A_K4me3 and Sample_B_K4me3 different patients?

[Researcher]

Yes these are different patients but enriched for same IP. Just to clarify, In the above example Sample A, B and C belong to subtype 1 while Sample 1, 2 and 3 are from subtype2.

[ATpoint]

Then these are biological replicates so exactly what you need for a meaningful analysis.

[Researcher]

Sample_A and Sample_B are independent samples from same tumor subtype. I just want to call consensus peak between them say for each IP and then compare it with the consensus peak from other subtype. Is it a good approach to move forward?

[Researcher]

@ATpoint, I am looking forward to your kind suggestion.

Thanks

Answer 1

[vj]

If you are looking for differential bound sites that are consistent then yes, in theory you can use them as replicates. If these are patient samples, then you can never have Patient A with Subtype 1 vs Patient A with Subtype 2 - all the more validation of why you used this approach. You can try using DiffBind which if I am not wrong uses read counts for user supplied peak regions and then uses DESeq/DESeq2 to evaluate differential binding.

[Researcher]

Thanks @vj for your reply. Yes thats true, I wouldn't have same patient's data from both the subtypes. Ok, If ill go with DiffBind, will that allow me to compare these unpaired samples from two different subtypes.

[vj]

Yes, you can. It has quite a generic design matrix, so you can define your replicates whichever way you wish. I would advice you to plot the tracks in UCSC to see if the significantly differentially bound regions are visually compelling as well.