Human protein-protein interactions are a key area of study when it comes to understanding the causes of many diseases like cancer. In order to analyze how proteins interact with each other, traditionally scientists use expensive experimental techniques or inefficient computational methods that scan through an entire search space. Recently, computational biologists have been able to tackle the problem of rigid body protein-protein docking using graph neural networks. In this thesis, we aim to improve the results of already existing protein docking machine learning models using in-house evolutionary data. We present a coevolution vector that encapsulates the evolutionary history between a pair of protein residues. By incorporating this vector as a node feature in the protein representation graph, we both guide existing models towards more accurate results and at the same time demonstrate the general utility of our coevolution feature in describing proteins.