3D genome assisted protein–protein interaction prediction

Protein–protein interactions (PPIs) play key roles in the biological system, making revealing protein interactome critical. A large portion of the protein interactome however, are yet to be revealed. Dozens of computational methods have been proposed to predict PPIs, but none of the existing method considers the DNA origin the interacting proteins in the perspective of 3D genome. In this work, we project the interacting proteins into the corresponding 3D genome loci and discovered that for the interacting protein pairs, their corresponding gene pairs are more likely to be proximate in chromatin 3D conformation. By incorporating 3D genome information, as well as adopting heterogeneous protein sequence encoding schemes, existing PPI prediction methods can be significantly improved in terms of accuracy and AUC. Combining our previous discoveries, we conjecture the existence of 3D genome driven cellular compartmentalization, which leads to an important principle that co-localized DNA elements can lead to increased probability of the co-localization of their downstream RNA and protein elements, thus facilitating the efficiency of the biological system.