Proteins are “social molecules.” Every protein in a living cell interacts with many other proteins and small-molecule ligands to carry out its biological function. The study of these protein interactions is fundamental for understanding biological processes for all organisms, in both healthy and diseased states. With advances in computational methods and in computer hardware resources, the ability to accurately and rapidly predict new protein interactions has emerged as an attractive approach and an excellent complement to experimental techniques. Methods have been developed by which we can study the interactions between proteins and other biomolecules by considering their overall sequence and structural similarities1,2. Nevertheless, it is well established that the overall sequence and fold similarities of proteins do not necessarily translate to similarities in protein function or to interactions between similar proteins3. It should be noted that the interactions of a protein with other proteins and with biomolecules primarily involve amino acids at the surface of a protein. This surface can be represented as a continuous surface that is imprinted with geometric and chemical features (“fingerprints”). In turn, it can be assumed that proteins engaging in similar interactions share common fingerprints on their surface representation4–6. Fingerprints may be difficult to grasp by visual analysis but could be learned from large-scale datasets using artificial intelligence.