Big data is a recurring problem in structural bioinformatics where even a single experimentally determined protein structure can contain several different interacting protein domains and often involves many tens of thousands of 3D atomic coordinates. If we consider all protein structures that have ever been solved, the immense structural space of protein-protein interactions needs to be organised systematically in order to make sense of the many functional and evolutionary relationships that exist between different protein families and their interactions. This article describes some new developments in Kbdock, a knowledge-based approach for classifying and annotating protein interactions at the protein domain level.