dc.contributor.author | Markowski, Etai | |
dc.date.accessioned | 2021-04-07T12:54:39Z | |
dc.date.available | 2021-04-07T12:54:39Z | |
dc.date.issued | 2021-04-07T12:54:39Z | |
dc.identifier.uri | http://hdl.handle.net/10222/80341 | |
dc.description | This thesis compares six phylogenetic tests, or equivalently, methods for constructing confidence sets of phylogenetic trees: the Kishino-Hasegawa (KH) test statistic, Shimodaira–Hasegawa (SH), two versions of the Approximately Unbiased (AU) test, Chi-square and Bonferroni. The Bonferroni test is a new variation of the Chi-square test that corrects for selection bias. A variation of the test (AU Corrected) is considered that adjusts for difficulties arising when bootstrap support for trees is low. Confidence regions for each test are examined using simulations from six and eight-taxon trees. We consider differing internal edge-lengths and challenging inference scenarios where some internal edge-lengths are equal to 0. Particular interest is in how the tests perform when a substantial amount of selection bias is expected such as when data are generated from a star tree, with multiple zero-length edges and with short internal edge-lengths. We notice that some tests such as SH are consistently conservative compared to all others. KH, while not designed to deal with selection bias, is surprisingly almost as conservative as SH. AU was similar in performance to but, by design, more conservative than AU Corrected. Their performance was poor with smaller internal edge lengths, Bonferroni offered a good compromise between being conservative and also closer from above to the simulation confidence level we expect it to have. The Chi-square test gave tight confidence sets but somtimes had coverage below the nominal confidence level. In the second part of this thesis we apply the same tests to multiple real-world data sets, some with larger numbers of taxa, and make references to the observations and trends obtained in the previous part. | en_US |
dc.description.abstract | This thesis compares six phylogenetic tests, or equivalently, methods for con- structing confidence sets of phylogenetic trees : the Kishino-Hasegawa (KH) test statistic, Shimodaira–Hasegawa (SH), two versions of the Approximately Unbiased (AU) test, Chi-square and Bonferroni. The Bonferroni test is a new variation of the Chi-square test that corrects for selection bias. A variation of the AU test, AU Corrected, is considered that adjusts for difficulties arising when bootstrap support for trees is low. Confidence regions for each test are ex- amined using simulations from six and eight-taxon trees. We consider differing internal edge-lengths and challenging inference scenarios where some internal edge-lengths are equal to 0. In the second part of this thesis we apply the same tests to multiple real-world data sets, some with larger numbers of taxa, and make references to the observations and trends obtained in the previous part. | en_US |
dc.language.iso | en | en_US |
dc.subject | phylogenetics | en_US |
dc.subject | ml | en_US |
dc.title | A COMPARISON OF METHODS FOR CONSTRUCTING CONFIDENCE SETS OF PHYLOGENETIC TREES USING MAXIMUM LIKELIHOOD | en_US |
dc.date.defence | 2021-03-26 | |
dc.contributor.department | Department of Mathematics & Statistics - Statistics Division | en_US |
dc.contributor.degree | Master of Science | en_US |
dc.contributor.external-examiner | n/a | en_US |
dc.contributor.graduate-coordinator | Joanna Mills Flemming | en_US |
dc.contributor.thesis-reader | Andrew Roger | en_US |
dc.contributor.thesis-reader | Hong Gu | en_US |
dc.contributor.thesis-supervisor | Edward Susko | en_US |
dc.contributor.ethics-approval | Not Applicable | en_US |
dc.contributor.manuscripts | Not Applicable | en_US |
dc.contributor.copyright-release | Not Applicable | en_US |