Protein-protein interactions (PPIs) are an essential part of many biological pathways in living organisms. With use cases such as regulation of gene expression, enzymatic catalyzation, and muscle contraction, understanding PPIs is a critical step toward a better understanding of life itself. Moreover, aberrant human PPIs may lead to multiple diseases, such as Alzheimer's, Creutzfeldt–Jakob, and cancer. Despite the undisputed importance of PPIs, only a small portion of the human interactome is known.

The PPI mapping problem is composed of two subproblems: the Interaction Problem—identifying the two or more proteins involved in a particular interaction; and the Position Problem—recognizing the residues within the interacting proteins that are crucial for the interaction (also known as hot spots or interacting residues). Current experimental techniques for PPI mapping, like Yeast 2 Hybrid or Alanine scans, are limited in scale, tedious, and expensive, therefore establishing the need for a fast, efficient, and accurate computational system.

DeepPPI is a Deep Learning algorithm that uses known PPIs to identify reoccurring patterns in the human interactome. These underlying patterns can be used, in turn, to predict both the existence of a new interaction and the interacting residues within the relevant proteins. Through this project, we hope to answer a fundamental biological question: How does nature, via evolution, create new protein-protein interactions? Additionally, we believe that DeepPPI will serve as a large-scale computational alternative to Alanine Scans and other experimental methods, contributing to the study of diseases and development of new therapeutics.