Luis Pedro Coelho: "Organizing the Proteome with Location and Function Topics"

Luis Pedro Coelho

Institute for Molecular Medicine, University of Lisbon, Portugal

Organizing the Proteome with Location and Function Topics

The subcellular location of a protein is one of the major characteristics of its behaviour. With automated microscopy and computer vision, we can study it at a proteome-wide scale.
Based on the Human Protein Atlas, a collection of fluorescently immuno-labeled proteins in human cells, we build a probabilistic graphical model which interprets the images in the context of functional annotations. The system automatically learns a small number of topics, based on location and function. Each protein is then assigned to a small number of these topics in a fractional manner. A large fraction of proteins are not located to single organelle, but are simultaneously present in multiple location, resulting in a mixture of pattern. Our topic modeling unmixes these patterns in location and function. The result is a better representation of the proteome than traditional assignment to organelles.
Finally, using projection into two dimensional space, we designed an interactive visualisation for exploring the proteome, where relationships between proteins become apparent.