The effect of running computeDeviations without Deviations

[JohnMCMa]

If I run computeDeviations on a chromVARCounts object without providing annotations, then I expect the deviations thus computed are the chromatin accessibilities of the input peaks. But is it safe to use such deviations to identify peaks that are differentially accessible between groups?

[AliciaSchep]

chromVAR was designed for applications where there are too few reads per peak for that kind of analysis to be meaningful, but I think it is reasonable to use the method for individual peaks in cases where coverage is high. When aggregating across peaks, the bias correction chromVAR does is important as biases can add up as you sum across peaks with similar properties; that bias correction may also be helpful for individual peaks. It would likely be a reasonable way of normalizing data for input into clustering or visualization, and for identifying high variance peaks. In terms of identifying differential peaks, we have not been rigorously compared this method with other methods designed expressly for that purpose. My suggestion would be to critically examine results....