A data-centric approach for investigation of protein-protein interfaces in Protein Data Bank

Understanding the structural architecture of protein interfaces is one of the key challenges in explaining how proteins interact and function. Protein-protein interfaces can be important targets for drug discovery and repurposing studies; this is only possible if the structural data available can be utilized in a meaningful way. Advancements in experimental techniques make it possible for scientists to determine larger and more complex protein structures. As these tools are becoming more accessible, the number of protein structures deposited in PDB is also increasing rapidly. This increase in the size of the protein complexes and the size of the data is an invaluable tool for understanding protein interactions. Still, it does not come without its computational challenges. This study first focuses on identifying protein-protein interfaces and overcoming some of the difficulties rooting from growing structure sizes in PDB. We evaluated 169,681 available protein structures in PDB and extracted 499,169 protein-protein interfaces. A unique set of protein interfaces can only be created by structural comparison of interfaces. The increasing number of interface structures makes it almost impossible to compare all interface structures in a reasonable time frame. Therefore, computational solutions are needed to reduce the comparison number within a manageable range without losing structural information. The second part of this study explains an approach for filtering highly similar proteins using sequential and structural information. As a result of this process, we identified 331,231 interfaces, unique in terms of sequence and structure. The reduced number of interfaces results in almost a 2.3-fold reduction in the number of comparisons needed, without losing any structural information. Implementing these techniques can make future studies on protein interface comparison possible for large data sets.