DiffBind without replicates

Question

[zhangdengwei]

Hi,

I am finding the differential peaks using DiffBind, but my sample has no replicates. And my SampleSheet is

> dbObj <- dba(sampleSheet="SampleSheet.csv")
trisomy_21 fibroblasts trisomy_21 trisomy_21 trisomy_21 1 narrow
euploid fibroblasts euploid euploid euploid 1 narrow
> dbObj
2 Samples, 33153 sites in matrix (47495 total):
          ID      Tissue     Factor  Condition  Treatment Replicate Caller Intervals
1 trisomy_21 fibroblasts trisomy_21 trisomy_21 trisomy_21         1 narrow     40820
2    euploid fibroblasts    euploid    euploid    euploid         1 narrow     44391

When performing the differential analysis, I got the following failure

> dbObj <- dba.contrast(dbObj, categories=DBA_FACTOR,minMembers = 1)
Error in dba.contrast(dbObj, categories = DBA_FACTOR, minMembers = 1) : 
  minMembers must be at least 2. Use of replicates strongly advised.
> dbObj <- dba.contrast(dbObj, categories=DBA_FACTOR,minMembers = 2)
Warning message:
No contrasts added. Perhaps try more categories, or lower value for minMembers. 
> dbObj <- dba.analyze(dbObj)
Error in pv.DBA(DBA, method, bSubControl, bFullLibrarySize, bTagwise = bTagwise,  : 
  Unable to perform analysis: no contrasts specified.
In addition: Warning message:
No contrasts added. Perhaps try more categories, or lower value for minMembers. 
> dbObj <- dba.contrast(dbObj, categories=DBA_CONDITION)
Warning message:
No contrasts added. Perhaps try more categories, or lower value for minMembers. 
> dbObj <- dba.contrast(dbObj, categories=DBA_CONDITION, minMembers = 1)
Error in dba.contrast(dbObj, categories = DBA_CONDITION, minMembers = 1) : 
  minMembers must be at least 2. Use of replicates strongly advised.

So, replication is requisite for DiffBind? If not, what can I do for this case? Thanks very much.

[sim.j.baum]

See answer from Devon Ryan below.
In the mean time while creating new data with replicates you could use bedtools to intersect regions between both ChIPseq exerpiemtns and identify regions mutually exclusive. Take the binding intensity differences into account. Validate those regions and differential binding for example by ChIP-qPCR - otherwise just do another ChIPseq experiment with replicates.

[ATpoint]

Hello zhangdengwei!

It appears that your post has been cross-posted to another site: https://support.bioconductor.org/p/125809/

This is typically not recommended as it runs the risk of annoying people in both communities.

[zhangdengwei]

So sorry! I would keep it in mind. Thanks for your reminder.

Answer 1

[Devon Ryan]

There are no statistically reliable methods that can be used to find differential peaks without replicates.

Answer 2

[zhangdengwei]

I got the answer from the developer of DiffBind. as follows

Yes, replicates are required to do any kind of statistical analysis. Replicates are required to estimate the variance in the data and calculate confidence statistics such as p-values/FDRs. Without replicates, you can do some exploratory analysis of overlapping peaks (occupancy analysis). For example using dba.plotVenn(). But not knowing if your data represents an outlier, combined with the inherent noisiness of peak calling, means you will have to have another way to validate any "differential" peaks you identify.