Hi Peter, please find below my response to your first email below (your second email arrived while I was writing). For paired-end reads, the insert size is known. Therefore, MEDIPS does not use the extend parameter but considers the entire fragment between mates for calculating coverage. If you process paired end data in single-end mode, information abt the insert size will be lost and reads will be extended by the extend parameter instead. Moreover, both mates of a Pair will be imported and considered for calculating coverage, therefore the counts are twice the actually sequenced DNA fragments. In my opinion, none of this has a severe negative effect on your results. However, processing paired-end data in paired-end mode is more appropriate. It still puzzle’s me what is going wrong in your case. Bam files are imported by Rsamtools and relevant parameters are: scanFlag = scanBamFlag(isPaired = T, isProperPair = TRUE, hasUnmappedMate = FALSE, isUnmappedQuery = F, isFirstMateRead = T, isSecondMateRead = F) Only the first mate of properly paired mates are supposed to be imported. As the first mate also has the information abt the insert size, MEDIPS will extend the first mate by the insert size prior to calculating coverage at genome wide windows. I am wondering, if e.g. your bwa paired-end bam files have ‘proper Paired’ reads included where both mates span long distances and have long insert sizes? All the best, Lukas Hi Peter, thank you for reporting this observation. Filtering of lowly covered genomic regions happens in the MEDIPS.diffMeth function which is called within the MEDIPS.meth function. Its really straight forward and I currently do not see what can go wrong here. Its roughly: (...) filter = rowSums(values) >= minRowSum (...) cat(paste("Execute edgeR for count data of", sum(filter), "windows...\n", sep = " “)) (...) d <- edgeR::DGEList(counts = values[filter, ], group = edRObj.group) (…) where values is a matrix with rows=genomic windows and columns=all samples at MSet1 and MSet2 (Input is not considered). Moreover, I do see an effect of the minRossum parameter in your example, i.e an increased value results in a decreased number of windows to be tested for differential enrichment. The major effect of reduced number of windows will be on the adjusted p-value. However, I also think its surprising that there remain so many windows which have a coverage > 1000 as a sum over all samples. Can you check if this is true by for example examining the count table for such windows? As you have only 6 ChIP samples, there must be ~166 reads per sample in average for windows surviving a minRowsum parameter of 1000. Do you have such a deep coverage? All the best, Lukas On 21 Apr 2016, at 18:53, peter.mcerlean [bioc] <noreply@bioconductor.org<mailto:noreply@bioconductor.org>> wrote: Activity on a post you are following on support.bioconductor.org<https: support.bioconductor.org=""/> User peter.mcerlean<https: support.bioconductor.org="" u="" 9754=""/> wrote Answer: MEDIPS: Number of windows tested in Paired-end data set<https: support.bioconductor.org="" p="" 81098="" #81117="">: Hi Lukas, I've just been running the analysis without the paired end parameters and have gotten better results: Unique MinRowSum Paired BWA Total Windows Windows Tested P<0.05 Adj P<0.05 1 50 FALSE FALSE 3,095,689 231,444 41,099 64 1 100 FALSE FALSE 3,095,689 65,927 18,429 183 1 200 FALSE FALSE 3,095,689 15,735 6,639 322 I've inspected some of the significant windows visually and are in agreement (i.e. Gain or loss in treatment Vs control). Can I trust these results over the paired-end approach? Thanks! P. ________________________________ Post tags: medips, r You may reply via email or visit A: MEDIPS: Number of windows tested in Paired-end data set