By MeV Bn Team (Raktim, Dan)

BNPredict Plugin

Bayesian Network (BN) analysis attempts to learn biologically meaningful gene interaction network (Directed Acyclical Graphs – DAGs) from of microarray data. The underlying assumption in this effort lies in the fact that most popular bioinformatics methods are good at identifying genes that discriminate between two or more groups or experimental conditions but often fails to elucidate the underlying mechanism of gene interaction that captures a biological process. This novel method uses seeds leaned from biomedical literature, protein-protein interactions or KEGG interactions or any combination to construct a starting ‘prior’ network. The machine learning algorithm then uses information from the microarray data to learn and refine the network constructed from prior information and predicts a high-confidence network. We also use bootstrap the samples to control for over fitting the network.

The BN module in MeV can be accessed from under the "Miscellaneous" category of algorithms. We provide annotations and support files for all major Affymetrix platform required to run this module and we are constantly adding new arrays and platform to our database.

Definitions

Bin(State): The biological expression levels of a gene with possible values Over-expressed/Up, Under-expressed/Down, and Unchanged/Neutral.

Perturbation: Changing the state of a gene.

Conditional Probability Table(CPT): A table indicating the likelihood of finding a gene in each of the three bins; over-expressed, under-expressed, and unchanged. This table is dependent on the states of all direct parent genes.

Bin(State) Probability Table: The probability table indicating the chances that a gene will be over-expressed, under-expressed, or unchanged. These values are calculated for each gene and are based on the set of all Conditional Probability Tables for every ancestor gene that lies upstream of a particular gene. Perturbations in the network have an effect on all the downstream BPTs.