A method for identification of parallelism in discrete character sets

It is proposed that parallel and reverse changes constitute the main factor governing the accuracy of phylogenies reconstructed from sets of discrete data. The relationship between such changes and Le Quesne's concept of compatibility, extended to multi-state characters, is examined and used as the basis of a computerized method, to estimate the proportion of those character states resulting from parallel changes. The results of applying the procedure to simulated protein sequence data and a variety of real data sets are presented, and these suggest that it is possible to estimate accurately (2%) the number of parallel character state changes represented in a data set, without the need to assume a phylogeny for the operational taxonomic units concerned. When real data sets are compared using this method it becomes clear that mammalian cytochrome-c is an exceptionally good protein for the reconstruction of tionary history, and that all other protein sets so far examined are likely to produce phylogenies which are significantly further from the 'best' phylo