search

biocore-ntnu / epic

(DEPRECATED) epic: diffuse domain ChIP-Seq caller based on SICER
http://bioepic.readthedocs.io
MIT License
31 stars 6 forks source link

epicR? #71

Open Bohdan-Khomtchouk opened 6 years ago

Bohdan-Khomtchouk commented 6 years ago

Any plans to create an R/Bioconductor package (e.g., epicR)?

endrebak commented 6 years ago

Thanks for the suggestion.

I almost never use R. What would the advantages be?

Since I do not do much R work, I do not think I would be the best person to do it anyways, but I'm open to the idea.

Bohdan-Khomtchouk commented 6 years ago

Mostly exposure to a very large and vibrant community called Bioconductor, which I believe would significantly boost epic's usage and popularity. epicR wouldn't be as much work as it may initially seem, due to R's fantastic reticulate package, which is used to great acclaim in R's API to TensorFlow (https://github.com/rstudio/tensorflow), amongst other places.

In my colleagues' experience, SICER has been historically difficult to setup/install and not particularly intuitive to run. Also, I've heard that the SICER algorithm is notoriously bad at overcalling false positives, at least relative to other peak callers like MACS2. I wonder how extensively epic has looked into these known issues.

endrebak commented 6 years ago

I did not implement the SICER algorithm that does not use input because in my experience it gives way too many false positives. Of course, as ChIP-Seq is 95% noise, it is hard to avoid this.

For H3K27me3 which I study, SICER seems to be the best one (I have tried thus far).

endrebak commented 6 years ago

The fact that SICER was hard to install and run was my main reason for rewriting it :) And also, the fact that it seems to give the best results on our data.

endrebak commented 6 years ago

I actually think this is a good idea from a "spreading the word" perspective, but it is not first on my todo-list.