(623h) Reverse Engineering the "Small-World" Gene Networks

The cDNA microarray data from experiments is used to arrange genes according to similarity in the gene expression ratios. This arrangement is in the form of subgroup or clusters, which are generally much smaller in number compared to the number of genes they represent, using various mathematical techniques. The average expression profiles of clusters generated can be used to model the genetic network with the help of reverse engineering techniques. The gene network gives an idea of existence and the level of interactions in-between the gene clusters, which can be put to use for many potential applications.

In this work, a robust clustering algorithm was improved upon and implemented using genetic algorithms with the jumping gene and siRNA adaptations. A multi-objective formulation of the algorithm was able to produce biologically relevant clusters for test data, and the work is in progress to get the results on real life data. Using the average cluster profiles, the reverse engineering of the gene networks was done using graph theoretical approaches. The approach exploits the fact that gene networks follow the ‘small world phenomenon’. The network was build up, starting from a single node, using fewer objectives and more constraints. This enabled us to solve a smaller dimensionality problem and solve the larger problem with a faster convergence rate. The following two models were proposed and implemented

1. Step by step optimization – uses the results of the n node problem as an initial seed to the GA of n+1 node problem
2. Step by step optimization with fixed previous topology – fixes the topology generated at the n node problem and the GA only optimizes the interaction of the (n+1)^th node, thus reducing number of variables

A set of Pareto optimal solutions were generated for each of the two cases which were in coherence with the available biological information.