WLoops: appropriate background model in scale-space representation of Hi-C maps improves detection of chromatin loops

Local enrichment methods to detect chromatin loops assign peaks in Hi-C maps as loops which are local maxima with respect to the neighboring regions and statistically higher than a local background model. Among the methods, a preferred approach called Mustache which relies on multi-scale theory has emerged that can detect chromatin loops at different scales. In this paper, we determine that the background model this widely used tool adopts to represent the Hi-C maps as not optimal. We show that by simply using the optimal background model (the Weibull distribution), we can improve substantially the detection of chromatin loops from Hi-C maps while keeping the general loop detection algorithm. Previously uncovered loops (in the order of thousands) can be detected with the new background model. These loops are supported by strong enrichment in Hi-C maps compared to local neighborhood. Biological confidence in these loops is observed because of overlap with binding of CTCF, and the Cohesin complex. Finally, we also find correspondence between these loops and interactions uncovered by other methods such as CHIA-PET and HiChIP. This study shows that while using favorable loop detection algorithm, adopting the wrong background model can disregard potential chromatin loops. The modified approach is made available as WLoops.