Different resolution produce different result about chess HOT 1 OPEN

Hi, sorry for the delayed reply. We've been working on refining an approach to select thresholds for SSIM and SN, as the thresholds of z-normalized similarity score ≤−1.2 and a signal-to-noise ratio of r ≥ 0.6 used in the original CHESS paper aren't appropriate for all datasets. Similarity and SN are influenced by the level of noise in the Hi-C matrices, so will vary across resolutions as higher resolutions are typically more noisy.

Our new approach is as follows: if you have biological replicates, and sufficient sequencing depth to analyse these individually, we suggest that you can use a comparison between biological replicates from the same condition to generate a "reference distribution" of SSIM and SN. You can then select thresholds based on this distribution, in order to identify regions that have lower SSIM and higher SN in your treatment-control comparison than you expect to find in the reference distribution.

Alternatively, if you don't have biological replicates but have access to another reference dataset, you can carry out a control-reference comparison and then take the difference of the SSIM profile between this and your real treatment-control comparison. This is an approach that was also used in the original CHESS paper for the Drosophila data.

We've added explanations of these approaches to the CHESS FAQ and they are also decsribed in a preprint. I hope this is helpful, please let us know if you have follow-up questions!

from chess.