Hi I have been using CHANCE (CHip-seq ANalytics and Confidence Estimation) to determine if my IP Efficiency is good after chIP-seq analysis. CHANCE has reported that for my cell line MIAPaCa-2 the IP Efficiency is weak for all my histone marks (in other words the IP failed); however, when I look at the bedgraphs that I outputted using HOMER I can see clearly defined peaks when I compare my unnormalized Histone mark bedgraph to the input. Additionally, when using other tools like HOMER to see IP efficiency it reports relatively good IP efficiency for at least the H3k27Ac and H3k4me3. I just wanted to get an idea if anyone else has had a similar problem and if so what did you do to optimize CHANCE ? Is there away to change how CHANCE trains its algorithm ?
Normally things like CHANCE work reasonably well for sharpish marks and become problematic when marks become broad. Anyway, there's no training in the CHANCE algorithm, it's effectively a modified Jensen-Shannon divergence. The only really tweakable things are the bins sizes and total number of bins that end up getting used. If you're using
plotFingerprint from deepTools to compute the CHANCE metric then you can change these. My guess is that you have very sparse ChIP datasets and very low-depth input, in which case the underlying algorithms are going to misbehave. In general, have a look at the plots produced by
plotFingerprint or similar tools and see if those match what you're seeing at the signal track level. If they don't, then don't put too much stock in the various QC metrics.