The genetic algorithm scheme for consensus sequences

A consensus sequence is a single sequence that represents characteristics of a family of sequences. Such synopses are most commonly used in the bioinformatics for sequence analysis. For example, algorithms that determine high quality consensus sequences are useful to construct a multiple alignment and consequently, a sequence logo (another representation that attempts to capture the important features of sequences). The determination of optimal consensus sequences is NP-hard (Gusfield). We present two new algorithms and compare them to earlier, published methods of determining consensus sequences. The first, CONSENSIZE, is an application of the genetic algorithm scheme (GAS). The other is a simple steepest descent search, usually not very useful for NP-hard problems, but surprisingly successful for this application. We discuss both algorithms and experimentally compare their accuracy and efficiency with the simulated annealing, multiple alignment and center string approaches. Test results are presented on both synthetic data and biological sequences.