# Chip-seq down-sampling problem

56234445 + 0 in total (QC-passed reads + QC-failed reads)
0 + 0 secondary
0 + 0 supplementary
0 + 0 duplicates
55173337 + 0 mapped (98.11% : N/A)
56234445 + 0 paired in sequencing
53808040 + 0 properly paired (95.69% : N/A)
54626508 + 0 with itself and mate mapped
546829 + 0 singletons (0.97% : N/A)
162718 + 0 with mate mapped to a different chr
112602 + 0 with mate mapped to a different chr (mapQ>=5)


IgG sample chip seq read stat

32745255 + 0 in total (QC-passed reads + QC-failed reads)
0 + 0 secondary
0 + 0 supplementary
0 + 0 duplicates
31546173 + 0 mapped (96.34% : N/A)
32745255 + 0 paired in sequencing
30934194 + 0 properly paired (94.47% : N/A)
31269484 + 0 with itself and mate mapped
276689 + 0 singletons (0.84% : N/A)
79786 + 0 with mate mapped to a different chr
49658 + 0 with mate mapped to a different chr (mapQ>=5)


knockout sample Rep 1 stat

28044731 + 0 in total (QC-passed reads + QC-failed reads)
0 + 0 secondary
0 + 0 supplementary
0 + 0 duplicates
27459222 + 0 mapped (97.91% : N/A)
28044731 + 0 paired in sequencing
26685254 + 0 properly paired (95.15% : N/A)
27162846 + 0 with itself and mate mapped
296376 + 0 singletons (1.06% : N/A)
116684 + 0 with mate mapped to a different chr
88965 + 0 with mate mapped to a different chr (mapQ>=5)


Knockout sample rep 2 stat

32410094 + 0 in total (QC-passed reads + QC-failed reads)
0 + 0 secondary
0 + 0 supplementary
0 + 0 duplicates
31774629 + 0 mapped (98.04% : N/A)
32410094 + 0 paired in sequencing
30802808 + 0 properly paired (95.04% : N/A)
31561218 + 0 with itself and mate mapped
213411 + 0 singletons (0.66% : N/A)
89688 + 0 with mate mapped to a different chr
63838 + 0 with mate mapped to a different chr (mapQ>=5)


Do i need to down-sample with respect to the IgG for control sample which has lot more reads at the same time Knockout sample 1 has less read. So is it good to go for further analysis or i need to down sample them ?

Any suggestion is really appreciated

• Usually downstream programs will take sequencing depth differences into account for you, so you only need to downsample if you get really funky results. Nov 20 '20 at 8:07
• "if you get really funky results." i did plot using your deeptools i do see a small peak in IgG so i feel there might be some issue.. which one would be better IgG or input im not clear .for the comparison ..any insight if IgG would be better or input ?
– kcm
Nov 20 '20 at 8:59
• control is almost always better than IgG. For this reason don’t bother with IgG. Nov 20 '20 at 17:59

No, downsampling in general has no use as it reduces power. Peak callers such as macs2 will internally scale data to account for sequencing depth differences.

The IgG itself is only useful for peak calling as it somewhat corrects for the unspecificity of the antibody to precipitate any kind of protein rather than the protein of interest. I hear people say, and personally agree with this by experience with my own experiments, that of course IgG pulls down notably fewer DNA that the target antibody, therefore the library requires notably PCR to produce enough material for sequencing. And this then makes it questionable how representative IgG is as a control. I personally prefer chromatin input to account for the intrinsic sequence/GC bias of the underlying genomic region in the sequencing/mapping. To my knowledge there are currently few established tools to really take these controls into account for downstream analysis such as differential testing via interaction terms because the library composition is simply so different from the IPs making normalization difficult.

Your focus imho should be rather to properly normalize your samples to correct for differences is signal/noise ratio, see for example my post over at biostars on how one could do that beyond naive per-million scaling.

• yes i seen that post even i wanted to use get count use deseq2 but the issue is it needs replicate and here one one of the replicate has less reads. I started with homer but after reading a bit shifted to macs " macs2 will internally scale data to account for sequencing depth differences." then i guess no need to downsample if i understand correctly and now I will go with input..
– kcm
Nov 20 '20 at 9:46
• Yes, if you use macs you do not need to do anything, the tool will correct depth internally. One remark, the link I shared is to visualize data for e.g. profile plots, it is not the input for peak calling. For this simply use the bam files as-is. Nov 20 '20 at 9:57
• i ran differential peak analysis using homer getdiffretiapeak function since only one replicate from ctrl and treatment worked..now is it advisable to run with one replicate and in the peak file argument ..so do i have merge both control and treatment peak and give or only one peak? i ran using merged both peak the result seems opposite..im not sure what is wrong..
– kcm
Nov 23 '20 at 22:00
• My advise is to run a proper experiment by doing replicates. I cannot comment on Homer since I use edgeR which is more reliable as it takes replicate information into account. ChIP-seq is so noisy (by personal experience) I would never do unreplicated ChIPs. Nov 24 '20 at 10:50
• "ChIP-seq is so noisy " im having the hard time RNA seq looks so easy now with compared to chip seq
– kcm
Nov 25 '20 at 11:30