Name:CALnet
Description:Combinatorial Association Logic Networks Inference
Abstract:Gene expression profiling of tumors has been widely used to reveal tumor characteristics. However, when retrieved from tumor tissue that has undergone the many steps of tumorigenesis it is virtually impossible to identify the initiating events that are causal for the development of the tumor. Therefore we have expression profiled 43 tumors that were induced by retroviral insertional mutagenesis, providing an excellent opportunity to study the association between the initiating events (the viral integration sites) and the consequent downstream expression profiles that result from tumor formation.
Multiple mutations are required to reach the tumorigenic state of uncontrolled cell proliferation. However, it is well known that these mutations are characterized by complex interactions. The most frequently occurring interactions are mutual exclusion (where one of multiple parallel alternatives may be selected) and cooperation (where all cooperating members need to be selected). For this reason, we expect associations between the insertion patterns and gene expression profiles to be more complex than a simple one-to-one correlation. To capture these complex associations, we infer small Boolean logic networks that explicitly incorporate operators to model the potential parallel alternatives (‘exclusive-or’ gates) as well as the potential cooperation between mutations (‘and’ gates). This requires solving a challenging optimization problem. To this end we employed a branch-and-bound algorithm tailored to account for the inherent risk of over-fitting.

Created:2010-05-01
Last updated:2010-05-01