This is a test version of Biostars. For the public version, visit https://www.biostars.org.
DiffBind - low number of significant peaks for one contrast

I am using Diffbind to call differential peaks on an ATAC seq dataset of four conditions (AW, BW, B, and C), and each condition has 2 replicates. One of my replicates (BW2) has low quality (low number of peaks detected by MACS2 compared to the other replicate, and low FRiP).

I got very low number of significant peaks for the comparison BW x AW, and it is expected because of the low quality of one of the replicates. But why I get a high number of peaks for the other comparisons that include this low quality replicate? I am worried if there is any issue with the analysis. Any suggestions?

dbObj <- dba(sampleSheet=samples, minOverlap=2)
consensus <- dba.peakset(dbObj, consensus = DBA_CONDITION, minOverlap=2)
consensus <- dba(consensus, consensus$masks$Consensus, minOverlap=1)
consensus.peaks <- dba.peakset(consensus, bRetrieve=TRUE)
counts <- dba.count(dbObj, peaks=consensus.peaks)
contrast <- dba.contrast(counts, categories=DBA_FACTOR, minMembers = 2)
analysed.consensus <- dba.analyze(contrast, method=DBA_ALL_METHODS, bBlacklist=FALSE, bGreylist=FALSE)

My results are:

Design: [~Factor] | 6 Contrasts:
  Factor Group Samples Group2 Samples2
1 Factor    BW       2     AW        2
2 Factor    BW       2      C        2
3 Factor    BW       2     BT        2
4 Factor    AW       2      C        2
5 Factor    AW       2     BT        2
6 Factor    BT       2      C        2

8 Samples, 113008 sites in matrix:
   ID Factor Condition Replicate    Reads FRiP
1 BW1     BW        BW         1 18351704 0.21
2 BW2     BW        BW         2 23909409 0.07
3 AW1     AW        AW         1 27899970 0.11
4 AW2     AW        AW         2 27712756 0.15
5  C1      C         C         1 19207760 0.44
6  C2      C         C         2 38438952 0.46
7 BT1     BT        BT         1 35313077 0.35
8 BT2     BT        BT         2 33108371 0.35

Design: [~Factor] | 6 Contrasts:
  Factor Group Samples Group2 Samples2 DB.edgeR DB.DESeq2
1 Factor    BW       2     AW        2      125         0
2 Factor    BW       2      C        2    83426     81864
3 Factor    BW       2     BT        2    54819     56533
4 Factor    AW       2      C        2    88260     84513
5 Factor    AW       2     BT        2    60529     61937
6 Factor    BT       2      C        2    25749     19873
diffbind atac
• 2,012 views • link
written by CB • CB

1 answer

I suspect that the answer lies in the FRiP values. The C and BT samples have consistently high FRiPs, indicating greater overall enrichment. It is likely that most consensus regions are much more lowly enriched in both the "good" BW sample and the poor one as compared to the C or BT samples. If the fold changes are great enough, even a low-powered experiment can detect the differences.

By this reasoning, the expected fold changes in the significantly different BW regions should overwhelmingly have the same sign (negative for contrasts 2 and 3).

• 0 views • link
written by Rory Stark Rory Stark

Thank you for your feedback.

• 0 views • link
written by CB • CB

Log in to answer this question.

More posts like this