Combining the original replicates and pseudo-replicates for differential binding analysis?

Question

[maisarasora]

Hi.

I want to compare the differential binding of suz12 (mixed peaks) in two conditions; Control(dmso) and Treatment(dox)

I have two replicates for each condition and used them for peak calling by EPIC2 against input

1. suz12dmso_R1
2. suz12dmso_R2
3. suz12dox_R1
4. suz12dox_R2

I checked the QC for the samples using ChIPQC, and it seems that Suz12dmso_R2 is pretty deviated from the suz12dmso_R1 and the rest of the samples. I tried to do the diffbind for the samples but no differential bound sites were obtained. I tried to merge the peaks of the replicates for each condition and used bedtools to check for the overlapping and unique sites between the two conditions. And there are sites that differentially bound by the peaks using the bedtools.

I'm still new in ChIp-seq analysis and diffbind. I'm not sure maybe because of suz12dmso_R2 is poorly correlated with the suz12dmso_R1, this cause diffbind cannot determine the differential sites.

I heard about pseudoreplicates and they usually used for IDR analysis. So I try to prepare the pseudoreplicates from the merged bam SUZ12DMSO and SUZ12DOX and prepare new bams files from half of the reads as additional samples for the replicates.

1. suz12dmso_pseudo1
2. suz12dmso_psedo2
3. suz12dox_pseudo1
4. suz12dox_pseudo2

I used them for peak calling using the same input files and parameters. At first, I tried to use the pseudoreplicates only for diffbind but analysis cannot be performed.

Next, I tried to combine all the samples both original replicates and pseudoreplicates for diffbind analysis. And I get significant differential sites for the two conditions. To confirm the results, I checked the sites against the bigwig files using deeptools for plotheatmap and the sites display different enrichment between the two conditions.

I don't know is my approach is legit or not. I hope I can get some opinions about this. Thanks!