gGN: Representing the Gene Ontology as low-rank Gaussian distributions

Alejandro A. Edera, Georgina Stegmayer, Diego H. Milone

Abstract: Highlights•A knowledge graph can be modeled using a set of Gaussian distributions efficiently.•Low-rank Gaussian distributions offer an accurate approach to graph representation.•This approach outperforms those using diagonal Gaussian distributions in accuracy.•A semantic-based loss function is proposed to learn low-rank representations.•The resulting representations are beneficial for semantic similarity analysis.