The most popular approaches for reconstructing phylogenetic trees attempt to solve NP-hard optimization criteria such as maximum parsimony (MP). Currently, the bestperforming heuristic for reconstructing MP trees is Recursive-Iterative DCM3 (Rec-I-DCM3), which uses a single tree (or solution) to guide its way through an exponentially-sized tree space. To improve performance further, we designed Cooperative Rec-I-DCM3, a population-based approach for utilizing a population of Rec-I-DCM3 trees.

more »

... compare the performance of Cooperative Rec-I-DCM3 to Rec-I-DCM3 on four large biological datasets. Of particular interest is Cooperative Rec-I-DCM3's selection criteria for maintaining a population of solutions. Our experiments reveal that diverse populations outperform Rec-I-DCM3 in terms of average rates of convergence to best-known MP scores. To achieve greater performance, we designed an elitist population strategy, in which each solution's tree score matches that of the best score found in each generation. The elitist strategy was by far the worst overall performer in our experiments. Hence, being greedy is not always the best approach. Instead, a population of diverse solutions allows our cooperative algorithm to achieve the greatest performance improvements.