|
Protein interactions are fundamental to all processes in the cell and are promising to identify disease genes. However, protein interaction networks are large, complex and of highly varying quality, which poses two challenging open problems:
1. How to predict disease genes despite noise in data and sparse networks? 2. How to reconstruct disease pathways and functional modules as causal models for disease and source of drug targets? Gregory Chaitin, one of the fathers of algorithmic information theory, argues that comprehension is compression. We introduce network compression as a novel approach to analyze protein interaction networks. Network compression identifies graph theoretic motifs and modules in networks reducing network complexity. We develop theory, algorithms, and applications to identify bio-markers and reconstruct disease pathways in Pancreas Cancer and in Parkinson's disease. |
|