Sushmita Roy, Terran Lane, Margaret Werner-Washburne
We present a probabilistic graphical modeling approach for condition-specific biological networks—networks of genes and proteins that exist in living cells under different environmental conditions. We model a biological network using Markov random fields and describe an algorithm for learning the structure of Markov random fields. We describe methods to evaluate a structure learning algorithm's ability to capture different dependencies. Unlike existing approaches for identifying condition-specific behaviour of networks, where condition-specific subnetworks are identified after learning separate networks per condition, we propose to simultaneously learn specific and generic subnetworks across different conditions. The structural and functional aspects of the condition-specific networks learned from different conditions will provide a holistic view of the cellular mechanisms employed to respond, and survive in stressful and healthy conditions.
Subjects: 12. Machine Learning and Discovery; 12.2 Scientific Discovery
Submitted: Apr 7, 2008