Identification of chromatin loops from Hi-C interaction matrices by CTCF-CTCF topology classification

Genome-wide profiling of long-range interactions has revealed that the CCCTC-Binding factor (CTCF) often anchors chromatin loops and is enriched at boundaries of the so-called Topologically Associating Domains or TADs, which suggests that CTCF is essential in the 3D organization of chromatin. However, the systematic topological classification of pairwise CTCF-CTCF interactions has not been yet explored. Here, we developed a computational pipeline able to classify all CTCF-CTCF pairs according to their chromatin interactions from Hi-C experiments. The interaction profiles of all CTCF-CTCF pairs were further structurally clustered using Self-Organizing Feature Maps (SOFM) and their functionality characterized by their epigenetic states. The resulting cluster were then input to a convolutional neural network aiming at the de novo detecting chromatin loops from Hi-C interaction matrices. Our new method, called LOOPbit, is able to automatically detect higher number of pairwise interactions with functional significance compared to other loop callers highlighting the link between chromatin structure and function.