Evolutionary biclustering based on expression patterns

The majority of the biclustering approaches for microarray data analysis use the Mean Squared Residue (MSR) as the main evaluation measure for guiding the heuristic. MSR has been proven to be inefficient to recognize several kind of interesting patterns for biclusters. Transposed Virtual Error (VE^t) has recently been discovered to overcome MSR drawbacks, being able to recognize shifting and/or scaling patterns. In this work we propose a parallel evolutionary biclustering algorithm which uses VE^t as the main part of the fitness function, which has been designed using the volume and overlapping as other objectives to optimize. The resulting algorithm has been tested on both synthetic and benchmark real data producing satisfactory results. These results has been compared to those of the most popular biclustering algorithm developed by Cheng and Church and based in the use of MSR.