tag:blogger.com,1999:blog-8386055846297828307.post6599515009482549007..comments2024-03-28T07:34:49.133+01:00Comments on The Genealogical World of Phylogenetic Networks: How to construct a consensus network from the output of a bayesian tree analysisUnknownnoreply@blogger.comBlogger5125tag:blogger.com,1999:blog-8386055846297828307.post-9082345811489311492013-08-16T17:10:02.898+02:002013-08-16T17:10:02.898+02:00Assuming your machine has enough RAM, there's ...Assuming your machine has enough RAM, there's no reason why R cannot process >10,000 trees. It just takes a bit of time, that's all. A job to leave running overnight perhaps. <br /><br />Maybe some benchmarking studies exist to see how this scales up with more trees/taxa etc?Rupert A. Collinshttps://www.blogger.com/profile/06603095778978682549noreply@blogger.comtag:blogger.com,1999:blog-8386055846297828307.post-72264011179079509802013-08-14T19:53:22.983+02:002013-08-14T19:53:22.983+02:00Thanks for telling us that, Rupert; it is not a pa...Thanks for telling us that, Rupert; it is not a package that I have had a chance to look at. Even 10,000 trees is quite a small sample for many people's datasets, and this is where reading all trees becomes problematic. DavidDavid Morrisonhttps://www.blogger.com/profile/05469392205239443608noreply@blogger.comtag:blogger.com,1999:blog-8386055846297828307.post-90421188974086732232013-08-14T16:47:15.789+02:002013-08-14T16:47:15.789+02:00Really helpful post. Thanks.
An alternative appro...Really helpful post. Thanks.<br /><br />An alternative approach is use the 'consensusNet' function in Klaus Schliep's R package 'phangorn'. Of the full 10,000 trees, I sampled 1,000 post-burnin trees randomly to calculate the networks. I repeated this a few times to make sure the results did not change drastically. <br /><br />It was a very easy procedure.Rupert A. Collinshttps://www.blogger.com/profile/06603095778978682549noreply@blogger.comtag:blogger.com,1999:blog-8386055846297828307.post-24910243540072918022013-08-14T14:59:12.782+02:002013-08-14T14:59:12.782+02:00Leonardo, Thanks for the clarification of an impor...Leonardo, Thanks for the clarification of an important point; and also for reading the blog. DavidDavid Morrisonhttps://www.blogger.com/profile/05469392205239443608noreply@blogger.comtag:blogger.com,1999:blog-8386055846297828307.post-78843020846093512682013-08-14T14:04:12.597+02:002013-08-14T14:04:12.597+02:00Actually the MAP tree is simply the most frequent ...Actually the MAP tree is simply the most frequent tree amongst the sampled ones, that is, the first one in the 'trprobs' file. The consensus tree, OTOH, is based on split frequencies. Both point estimates try to minimize the Bayesian risk, but based on distinct loss functions (e.g. <a href="http://sysbio.oxfordjournals.org/content/60/4/528" rel="nofollow">http://sysbio.oxfordjournals.org/content/60/4/528</a>). I don't see them as less or more Bayesian, but I understand your point that a network can represent much more information than a single tree -- and the beauty of a Bayesian analysis is in not neglecting the uncertainty ;)<br /><br />Thanks for the posts, by the way, I really enjoy this blog.Leonardo de Oliveira Martinshttps://www.blogger.com/profile/16384711695752944768noreply@blogger.com