Himmelstein's PhD research began with the question of how to teach a computer biology, leading him to explore heterogeneous networks (HetNets), which are labelled property graphs. He created a graph with over 2 million relationships using Neo4j and integrated knowledge from public resources, resulting in a HetNet called Hetionet Version 1.0. This network contains 50,000 nodes of 11 types and 2.25 million relationships of 24 types, mined for drug repurposing by identifying patterns in paths that are predictive of treatment or efficacy. The project, called Rephetio, uses machine learning to predict the probability of treatment for all 200,000 compound-disease pairs, with notable successes including predicting bupropion's effectiveness against nicotine dependence. Himmelstein's work showcases the potential of HetNets and Neo4j in biomedical research and drug discovery.