tariks/peakachu

diffPeakachu with newer models

ViriyaK opened this issue · 4 comments

ViriyaK commented

Hello,

I was trying to use diffPeakachu to find differential loops between two conditions but I see that pair-probs.py is currently specifically mentioning the CTCF and H3K27ac trained models. I had called loops using the new high confidence models and wondering how I should proceed.

Also, your links to some of the models from github currently does not work.

Thanks!

Hendricks27 commented

I had the same question here actually...

XiaoTaoWang commented

Hi all, I have updated the README file, so that all links work. However, the script under the diffPeakachu folder was uploaded only for reproducing a figure in our original paper, and is not compatible with newer models trained with peakachu version >=2.0. Sorry for the inconvenience.

XiaoTaoWang commented

There are actually many ways to identify differential loops based on peakachu predictions. For example, a pixel can be defined as a differential loop if its peakachu score in one sample is greater than 0.9, but lower than 0.5 in the other sample.

Hendricks27 commented

Hey Dr. Wang,

Thank you so much for your detailed reply. It definitely clarifies a lot!
For differential loop calling, is there a performance difference between the threshold method (which you just mentioned) and the gaussian mixture model?
Thanks!