Goutham- The first "peakset" correlation heatmap is heavily dependant on how peaks were called, as samples that don't have a specific peak called have no score for that peak. The second "count" heatmap gives a much less biased view of how samples, including replicates, are correlated, as it takes into account the read density at every site in every sample. If your replicates cluster well in the count correlation heatmap, that is a good sign, but it may be worth trying to understand why they were so divergent at the peak-calling level. Some venn diagrams and browser views may help in that. Be sure to look at the controls! Cheers- Rory From: Goutham atla <goutham.atla@gmail.com<mailto:goutham.atla@gmail.com>> Date: Mon, 27 Jan 2014 18:30:49 +0530 To: Rory Stark <rory.stark@cruk.cam.ac.uk<mailto:rory.stark@cruk.cam.ac.uk>> Subject: DiffBind Questions Dear Rory, We have been using DiffBind on our Chip-seq data. I would like to know, to see the correlation replications.... There are two ways to generate heatmaps between replicates...one is intially loading the samplesheet. roughly: NE=dba("sampleSheet.csv") plot(NE) The heatmap generated by the above method is showing low correlation values. But the heatmap generated after dba.count(), is showing good correlation. Now to understand the correlation between replicates, which heatmap should we consider ? Thanks and Regards, -- Goutham Atla [http://pdbparser.com/images/1380912429_social_twitter_box_blue.png]<h ttps:="" twitter.com="" geek_y="">[http://pdbparser.com/images/1380912469_soci al_linkedin_box_blue.png]<http: www.linkedin.com="" pub="" goutham-="" atla="" 22="" 698="" 974=""> [[alternative HTML version deleted]]