Hello,

Rory Stark I have a question regarding the 3 scenarios in which lfcShrink is applied in DiffBind package. The source code looks like this

 if(shrink) {
if(contrast$contrastType=="byname") {
  res$de <- suppressMessages(DESeq2::lfcShrink(pv$DESeq2$DEdata,
                                               coef=contrast$contrast,
                                               lfcThreshold=lfc,
                                               res=res$de,type="apeglm"))
} else  if(contrast$contrastType=="results1") {
  res$de <- suppressMessages(DESeq2::lfcShrink(pv$DESeq2$DEdata,
                                               coef=contrast$contrast[[1]],
                                               lfcThreshold=lfc,
                                               res=res$de,type="apeglm"))
} else {
  res$de <- suppressMessages(DESeq2::lfcShrink(pv$DESeq2$DEdata,
                                               contrast=contrast$contrast,
                                               lfcThreshold=lfc,
                                               res=res$de,type="ashr"))
}

In the above code, what do contrast$contrastType=="byname" and contrast$contrastType=="results1" mean? Could you explain what these three scenarios are?

Further, can the lfcShrink function be turned off so that fold values are calculated as condition1 - condition2 for DESeq2? Is there a scenario where that is possible in DiffBind? I deeply appreciate your help and suggestions. Thank you so much!

Priyanka

You are looking at some fairly deep DiffBind internals here. The contrastType has to do with how the contrast was specified when calling 'dba.contrast()' (by coefficient, by factor values, etc) See the help page where it discussed the contrast parameter; if you really want to dive in, these values are set by the DiffBind:::pv.checkContrast() function.

DiffBind uses shrunken fold change calculations as this is considered best practice. You can however easily compute your own fold changes. For example:

data(tamoxifen_analysis)
dbreport <- dba.report(tamoxifen)
names(mcols(dbreport))
dbreport$Fold <- dbreport$Conc_Resistant - dbreport$Conc_Responsive
dbreport

Note that even though these are not "shrunken", they may still be of small magnitude.