As an extension to the already published method, we have also developed a Cytoscape plug-in that does predictive modeling of the BN described above. For each network loaded with an associated conditional probability table (CPT), the plugin calculates probabilities of expression (bin/state probabilities) for each network node. For each node, the CPT is used to determine a specific probability table that is not dependent on its parent node(s). When a network is perturbed using BNPredict, the set of CPTs for the network is updated to reflect the set of conditions applied to the system. From this updated set of CPTs, a new collection of bin probability tables are calculated. By examining the expression patterns and the changes to the expression patterns as a result of perturbations it is possible to predict network reactions to specific stimulus.
In a nutshell, it attempts to predict the state of gene A given the state of its parent Gene B. The possible states a gene can exist in are: up regulated, down regulated or unchanged. Once a network has been leaned we find the conditional probability table (CPT) associated with each node (gene). The CPT of a node constitutes the individual probabilities of the gene being up, down or unchanged given its parent(s) is/are in state up, down, unchanged (any combination of parent and state). Knowing the CPT of any node we can then predict the exact probability of the node being in any one state given any combination of parents and their states.
This method allows researchers to conditionally alter gene expression and predict resulting changes in a biological process based on microarray data which can then be experimentally validated. We hope this novel approach would enable researchers to (a) get a better understanding of the biological interactions that exists in an enriched set of genes and (b) also to predict yet unknown processes and/or interactions. 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.