Spatial Neighbor Models Applied to Genetic Regulatory Network Inference
David S Burton,Matthew Nicholson McCall,Tanzy Love University of Rochester
Abstract
Though a wealth of gene expression data is now available, statistical methods which are able to model that data to infer network structure from gene product interactions are still lacking. We propose the application of spatial statistics models to infer network structure and functionality from gene expression datasets. Spatial models have several features which can be used to represent common network mechanisms such as cyclical behavior, decay of signal, mediated effects through parent nodes, as well as directionality of effect and parent-child relationships. We use a hierarchical Bayesian sampling method to estimate parameters. By comparing model fitting diagnostics among many candidate network structures, the model which best fits the data can be revealed, as well as a grouping of models within an interval of the best model fitting score. Results are compared on datasets which have been used to estimate networks using other inference methods.
Keywords: networks,gene expression,spatial models