THE DISCOVERY AND IMPORTANCE OF MULTIPLE ISLANDS OF MOST-PARSIMONIOUS TREES

The set of most-parsimonious trees for a data matrix may include several distinct classes (islands) of trees. An island is defined as a collection of trees, all less than a specified length, each tree connected to every other tree in the island through a series of trees, and each one differing from the next by a single rearrangement of branches. One advantage of this method of defining classes of trees is that islands are easily found by tree-searching programs. Multiple islands can be discovered by conducting many searches using a tree-searching program, each search beginning with a different tree. Among 37 data matrices examined, all 8 data matrices with two or more islands have retention indices less than 0.67. Trees are generally more similar within islands than among islands, as judged by analysis of partition metric distances between trees. As a consequence, trees in different islands may have different implications for character evolution, and for this reason should be sought.