tag:blogger.com,1999:blog-7785493.post7075683492000848673..comments2024-01-04T04:11:55.717+02:00Comments on Dienekes’ Anthropology Blog: Clusters galore in HGDP panelDienekeshttp://www.blogger.com/profile/02082684850093948970noreply@blogger.comBlogger13125tag:blogger.com,1999:blog-7785493.post-32601013393955268052013-09-05T18:48:37.860+03:002013-09-05T18:48:37.860+03:00How do you decide on the maximum number of cluster...<i>How do you decide on the maximum number of clusters to consider in mclust (i.e., where does maxclust=70 come from)?<br /></i><br /><br />The number of inferred clusters should be lower than maxclust. So, if you set, e.g., maxclust=50, then MCLUST will give you a warning that the number of clusters occurs at Max choice, and thus it will be a good idea to increase maxclust and re-run it.<br /><br />You should make maxclust as high as your computing power allows. Dienekeshttps://www.blogger.com/profile/02082684850093948970noreply@blogger.comtag:blogger.com,1999:blog-7785493.post-8223335622569574222013-09-05T17:26:17.678+03:002013-09-05T17:26:17.678+03:00How do you decide on the maximum number of cluster...How do you decide on the maximum number of clusters to consider in mclust (i.e., where does maxclust=70 come from)?Anonymoushttps://www.blogger.com/profile/10260930454565449628noreply@blogger.comtag:blogger.com,1999:blog-7785493.post-9414385808313741872011-04-03T14:55:04.139+03:002011-04-03T14:55:04.139+03:00You are right.You are right.Dienekeshttps://www.blogger.com/profile/02082684850093948970noreply@blogger.comtag:blogger.com,1999:blog-7785493.post-85587810355791874232011-04-03T14:31:18.905+03:002011-04-03T14:31:18.905+03:00I have no experience of R or mclust, but I think t...I have no experience of R or mclust, but I think there may be an error in the instructions. I get an error that Z is not found.<br /><br />Should<br />> MDS <- read.table("plink.mds", header=T)<br />> maxclust <- 70<br />> MCLUST <- Mclust(Z[, 4:(d+3)], G=1:maxclust)<br /><br />Be<br /><br />> Z <- read.table("plink.mds", header=T)<br />> maxclust <- 70<br />> MCLUST <- Mclust(Z[, 4:(d+3)], G=1:maxclust)<br /><br />?<br /><br />GarvanGarvanhttps://www.blogger.com/profile/08917487267933584193noreply@blogger.comtag:blogger.com,1999:blog-7785493.post-89660721425385014522011-03-14T20:30:18.548+02:002011-03-14T20:30:18.548+02:00You can cluster a large number of populations inst...You can cluster a large number of populations instead of a large number of individuals.Dienekeshttps://www.blogger.com/profile/02082684850093948970noreply@blogger.comtag:blogger.com,1999:blog-7785493.post-80923664383773608472011-03-14T19:03:02.052+02:002011-03-14T19:03:02.052+02:00Dear Dienekes,
Thank you for your quickly answer.
...Dear Dienekes,<br />Thank you for your quickly answer.<br />Do you think that I could use it with population data instead that individual?bauhttps://www.blogger.com/profile/13358664543737897087noreply@blogger.comtag:blogger.com,1999:blog-7785493.post-53449055314326969142011-03-14T18:55:44.564+02:002011-03-14T18:55:44.564+02:00Dear Dienekes,Thank you for your answer. Do you Th...Dear Dienekes,Thank you for your answer. Do you Think that i could use this method to define two (or more)cluster starting from population data instead individual data?PuDuhttps://www.blogger.com/profile/12100840021963415248noreply@blogger.comtag:blogger.com,1999:blog-7785493.post-39169241090975871652011-03-10T22:30:01.670+02:002011-03-10T22:30:01.670+02:00I am not familiar with the way adegenet outputs it...I am not familiar with the way adegenet outputs its MDS representation. <br /><br />The MDS object in the given code is a N by D array where:<br /><br />N: number of individuals<br />D: number of MDS dimensions<br /><br />As long as you are able to put your data in this format, it doesn't matter which software produces the MDS or PCA representation.Dienekeshttps://www.blogger.com/profile/02082684850093948970noreply@blogger.comtag:blogger.com,1999:blog-7785493.post-23672120261756987552011-03-10T17:17:46.666+02:002011-03-10T17:17:46.666+02:00Dear dienekes,thank you for your job.
Could i make...Dear dienekes,thank you for your job.<br />Could i make the same analysis starting from an mds made with adegenet?<br />Thank you for your patiencebauhttps://www.blogger.com/profile/13358664543737897087noreply@blogger.comtag:blogger.com,1999:blog-7785493.post-961057260398894452010-11-30T11:17:14.617+02:002010-11-30T11:17:14.617+02:00Why were the Maasai and Ethiopians removed?
They ...<i>Why were the Maasai and Ethiopians removed?</i><br /><br />They were not, because they are not in the HGDP panel to begin with.Dienekeshttps://www.blogger.com/profile/02082684850093948970noreply@blogger.comtag:blogger.com,1999:blog-7785493.post-59754696302632642942010-11-30T11:14:26.463+02:002010-11-30T11:14:26.463+02:00Why were the Maasai and Ethiopians removed?Why were the Maasai and Ethiopians removed?Lankhttps://www.blogger.com/profile/09164328821211694856noreply@blogger.comtag:blogger.com,1999:blog-7785493.post-27874749842547475082010-11-30T03:26:31.250+02:002010-11-30T03:26:31.250+02:00"This is rather astonishing. There are many c..."This is rather astonishing. There are many clusters with 100% correspondence to HGDP populations. A few populations, mostly from regions with high levels of inbreeding are split into multiple sub-clusters, perhaps reflecting some type of tribal affinity".<br /><br />Uh? After one removes too low samples (I used n=10 as treshold), the mono-clustering populations are just 17, while the multi-clustering ones are 26. Total: 43. Mono-clustering populations represent only a bare 40% of all relevant samples. <br /><br />Furthermore, some of the mono-cluster populations are clearly "tribal" and often alleged as "inbreeding", such as, in Europe: Basques, Sardinians, North Italians, Tuscans... but not French, Russian, Adigey nor Orcadian, all clear cases of multiple origins. <br /><br />Valid homogeneous clusters are also found in Africa when the populations are homogeneous (the two Pygmy groups, Mandenka, Yoruba), among some rather isolated East Asian groups (Uygur, Dai, She, Tu, Japanese), but not among more cosmopolitan groups such as Han, Mongola, Cambodians or Yakuts (at least two population layers at their origins). <br /><br />[No follow up. If you really need to contact me, follow my profile].Majuhttps://www.blogger.com/profile/12369840391933337204noreply@blogger.comtag:blogger.com,1999:blog-7785493.post-19568736556154808132010-11-30T02:24:57.851+02:002010-11-30T02:24:57.851+02:00North Italians and Tuscans are on the same line, b...North Italians and Tuscans are on the same line, but also 2 French on 28. We know that more than four millions French are of Italian extraction. If the sample is representative, they would be 7%. What a pity that there aren’t Spaniards, otherwise it would be interesting to see their location.Gioiellohttps://www.blogger.com/profile/13578860964923773647noreply@blogger.com