Vous êtes iciPublications

NoiseSnapper: A Method for Detecting Noise in Phylogenetic Analysis


Par Cyril - Posté le 17 avril 2014

TitreNoiseSnapper: A Method for Detecting Noise in Phylogenetic Analysis
Type de publicationConference Proceedings
Year of Conference2002
AuteursZaragüeta i Bagils R, Gallut C, Vignes Lebbe R
Nom du type21st Annual Meeting of the Willi Hennig Society
Volume19
Pagination162
Année de publicationAugust
Année de publicationCladistics
Conference LocationHelsinki
Résumé

Cladistic analysis is a method for choosing the cladogram that has the greatest explanatory power, i.e., which minimizes ad hoc hypotheses. Characters are thus considered either as corroborations of the chosen cladogram or as refutations of other hypotheses. To be considered as potential tests, characters must bear some information. NoiseSnapper is a test for the null hypothesis of the absence of phylogenetic information content of a character in a particular context. In order to evaluate a character's information content, NoiseSnapper generates a series of a 1000 random characters; within each of these matrices all the generated characters have the same number of states. The homoplasy content for each of the randomly generated characters is measured with the cladogram obtained with the real characters, by means of its RI (retention index) value. The distribution of these index values is used to define a threshold value of fit. The null hypothesis is rejected for a real character if it is less homoplastic than 95% of the random characters. Because, on average, a random character fits as well any cladogram, if the quantity of homoplasy beard by a real character is, at least, as great as the threshold value defined, then it cannot corroborate nor refute the cladogram and it cannot be used as an argumentation to defend the hypothesis found. A character is considered ''random'' in a particular context, that is, measuring its homoplasy with a particular cladogram that corresponds to a particular taxonomic sampling. It does not mean that the character is intrinsically uninformative. NoiseSnapper also increases the precision of parsimony, showing that some of the most parsimonious trees (MPTs) can have different information contents. It can be used to choose one among these MPTs. We discuss the criteria used in this choice in terms of the number of random characters and RI values. As NoiseSnapper is a method that detects noise on cladograms, eliminating random characters in a new analysis usually does not modify the cladogram. Because noise tends to weaken the argumentation, the argumentation will be stronger. NoiseSnapper can also be used to explore which characters support a node that was not expected or one that is not accepted in a particular framework. It can also be used to reassess character definition, i.e., a random character might be based on illdefined homology hypotheses and thereby should be redefined.