Bayesian Interaction and Association Networks From Multiple Replicates of Sparse Time-Course Data

Biological experiments of proteins and genes often involve the collection of multiple replicates of sparse time-course data. From such time-course data, protein (or gene) interaction posterior probabilities are computed based on individual and multiple replicates. This is accomplished through Bayesian inference in conjunction with the Metropolis-Hastings algorithm. The Bayesian posterior probability is computed for two distinct cases. One case assumes the replicates are independent events, the other assumes the replicates are not independent events (using a hierarchical structure). Closed form Bayes factors are developed for each situation. In order to test the algorithm's ability to identify signal, multiple replicates of simulated network data are generated and modeled. Two biological data sets, Arabidopsis thaliana and PC-3, are also modeled, each consisting of multiple replicates. For multiple replicates, modeling is done in accordance with the afore mentioned independence and non-independence assumptions among replicates. Models are also produced for individual replicates. Our algorithms produce high protein (or gene) interaction posterior probabilities to pairs of proteins when they have at least moderate partial correlation.

Grantor dc:publisher

Wake Forest University

Year dc:date.issued

2012

Author dc:creator

• Patton, Kristopher Laurence

• Bayesian

Language dc:language.iso

en

Handle dc:identifier.uri

http://hdl.handle.net/10339/37257

OAI identifier oai:identifier

oai:wakespace.lib.wfu.edu:10339/37257