Bayesian methods for elucidating genetic regulatory networks
A popular method for inferring gene regulatory networks from time series data uses Dynamic Bayesian Networks (DBN).DBN methods estimate a probabilistic graphical model, given the time-series data.Often this takes the form of a model selection problem, and methods such as the Least Absolute Shrinkage and Selection Operator (LASSO), but usually for steady-state rather than time-series data. Recently, mutual information methods have been extended to analyze time-series data and produce directed networks[ We also carried out extensive empirical studies of our new method.
The first dataset measures the gene expression levels over time of 97 yeast segregants perturbed with the drug rapamycin.We focus on a few that compare the inferred network with a gold-standard network of true edges.One measure that we use is the precision of the inferred network, equal to the number of true positives divided by the total number of edges in the inferred network.Specifically, Scan BMA generally produced more favorable areas under the Receiver-Operating Characteristic and Precision-Recall curves than other regression-based methods and mutual-information based methods.In addition, Scan BMA is competitive with other network inference methods in terms of running time.
We compared Scan BMA to other popular methods using time series yeast data as well as time-series simulated data from the DREAM competition.