Machine learning could be a powerful new tool in the study of protein interactions. Computational biologist Bruno Correia at École Polytechnique Fédérale de Lausanne, has designed a new system for “molecular surface interaction fingerprinting” (MaSIF), which utilizes a special area of machine learning called geometric deep learning to analyze the outer surfaces of proteins and predict how they might interact with other proteins. Like modern facial recognition software, MaSIF is trained on a large set of data to recognize recurring patterns. MaSIF starts with some basic information about the physical curvature, electric charge, and hydrophobic characteristics of protein surfaces, and is then trained to recognize when these features combine into higher-level patterns. Different versions of MaSIF boast impressive achievements in terms of accuracy and speed, when compared to similar existing systems. You can read more about this fascinating technology at Quanta Magazine.