How to evaluate the detected loops? about chromosight HOT 4 CLOSED

Hi Siyuan, thanks for your interest in Chromosight. The detections are established by calculating the correlation (Pearson coefficient) with a generic loop pattern.
The patterns are said to be positive detections if the computed coefficient is above a threshold (by default, we put the value 0.4 but you can change it with the option -pearson).
Then the pvalue are computed with the function corr_to_pval: given a list of Pearson correlation coefficient, the function converts them to two-sided log10 p-values. The p-values are computed via the fisher transformation described on: https://w.wiki/Ksu

link to the function: corr_to_pval
Hope it helps.

from chromosight.

Thank you for the prompt response!

To be more specific, does that mean that the Pearson coefficient is calculated by comparing the output "prefix.tsv" with the ground truth file?

Thank you again for your advice!

from chromosight.

The Pearson coefficient is computed between patches of your contact data (for example patches of 17x17 along your normalized contact map) and a generic pattern for example a loop pattern 17x17 (that we created from experimental datasets or can be generated with a gaussian 2D kernel)

Files for generic patterns of loops, borders, hairpins can be found here:
https://github.com/koszullab/chromosight/tree/master/chromosight/kernels
Best,

from chromosight.

Thank you very much for your advice!

I will close this issue.

from chromosight.