Where do these kernels come from?

[zhyanlin]

Where do these kernels come from?

[axelcournac]

Hello,
Thanks for the feedback. The kernel loop were initially created with Hi-C data from yeast S. Cerevisiae at 2 kb resolution but it can apply to various different types of data. There are 2 different loop kernels currently implemented that you can choose with the kernel option: loops and loops_small that have a different size. You can make your own kernel (for example a simple theoretical 2D gaussian filter) by creating your own json file:
https://chromosight.readthedocs.io/en/latest/TUTORIAL.html#generating-custom-patterns

[zhyanlin]

Thank you!

[zhyanlin]

The kernel loop were initially created with Hi-C data from yeast S. Cerevisiae at 2 kb resolution

I realized that I still didn't understand how to get the loop kernel.
I assume that loops in the yeast S. Cerevisiae contact map are unknown until applying your kernel to detect them. Did you annotate loops in yeast with another tool or use experiments such as ChIA-PET to get loops to define kernels?

[axelcournac]

We extracted by eye about twenty loops that were very clear in S. Cerevisiae and performed the pileup. This already gave a satisfactory kernel, which could also be corrected by hand if there was still some noise or to make it symmetrical. We did not use an alternative technique like ChIA-PET for the first annotation.

[zhyanlin]

Thank you for your answer!