INFERENCE OF FUNCTIONAL NETWORKS OF CONDITION-SPECIFIC RESPONSE - A CASE STUDY OF QUIESCENCE IN YEAST

SUSHMITA ROY, TERRAN LANE, MARGARET WERNER-WASHBURNE, DIEGO MARTINEZ
2008 Biocomputing 2009  
Analysis of condition-specific behavior under stressful environmental conditions can provide insight into mechanisms causing different healthy and diseased cellular states. Functional networks (edges representing statistical dependencies) inferred from condition-specific expression data can provide fine-grained, network level information about conserved and specific behavior across different conditions. In this paper, we examine novel microarray compendia measuring gene expression from two
more » ... e stationary phase yeast cell populations, quiescent and non-quiescent. We make the following contributions: (a) develop a new algorithm to infer functional networks modeled as undirected probabilistic graphical models, Markov random fields, (b) infer functional networks for quiescent, non-quiescent cells and exponential cells, and (c) compare the inferred networks to identify processes common and different across these cells. We found that both non-quiescent and exponential cells have more gene ontology enrichment than quiescent cells. The exponential cells share more processes with non-quiescent than with quiescent, highlighting the novel and relatively under-studied characteristics of quiescent cells. Analysis of inferred subgraphs identified processes enriched in both quiescent and non-quiescent cells as well processes specific to each cell type. Finally, SNF1, which is crucial for quiescence, occurs exclusively among quiescent network hubs, while non-quiescent network hubs are enriched in human disease causing homologs. 12:58 Proceedings Trim Size: 9in x 6in qnq˙mbs behavior typically identify genes differentially expressed across conditions. Fine-grained, interaction analysis among differentially expressed genes can provide deeper insight into human diseases such as cancers 1 . We define functional networks as networks with edges representing general, statistical dependencies among genes. We model functional networks using undirected, probabilistic graphical models, Markov random fields (MRFs). We present a new algorithm, Markov blanket search (MBS), for learning the MRF structure. MBS is based on Abbeel et al.'s theoretical work of Markov blanket canonical parameterization (MBCP) 2 , which requires an exhaustive enumeration (O(n l )) over variable subsets up to size l, where n is the number of variables. We establish an equivalence between the MB canonical parameters and per-variable canonical parameters 3 , which requires enumeration over only singleton sets (O(n)), thus providing a tractable approach for learning genome-scale networks. We apply our algorithm to two novel yeast (S. cerevisiae) microarray datasets measuring gene expression of quiescent and non-quiescent cells, isolated from glucose-starved stationary-phase cultures 4 . Quiescent cells play important roles in health and disease conditions of most living systems, but have been difficult to study due to their low metabolic activity 5 . The recent generation of microarray datasets for these cells 4 , gives us the first opportunity to infer a functional network that provides a fine-grained characterization of yeast quiescence. Algorithms for functional network inference can be broadly classified into: (a) pairwise models (capturing dependencies between only pairs of nodes), and (b) higher-order models (capturing dependencies among two or more nodes). Bayesian networks are higher-order, directed models 6,7,8 , but the acyclic constraint of the graph structure cannot easily capture cyclic dependencies. Undirected graphical models can represent cyclic dependencies, but because network inference is much harder 2 , higher-order dependencies are often approximated by lower-order (often pairwise) functions 9 . As biological networks are likely to have higher-order dependencies 10 , higher-order models are more appropriate for modeling functional networks. Unlike Bayesian networks, our model captures cycles and, unlike pairwise models, we explicitly identify higher-order dependencies. Bio-techniques for identifying condition-specific networks have been applied to transcriptional regulatory networks 11 . Computational identification of functional networks can provide a less expensive, complementary view of condition-specific networks. Some existing computational approaches integrate mRNA expression with known protein networks 1 . Other
doi:10.1142/9789812836939_0006 fatcat:ueeapm3z5jazxl5kfwhlrljvxa