[R-sig-phylo] Some questions about pPCA
Dear all, I'm a new subscriber to this list since I just started to play with phylogenetic data with R. The task is facilitated by reading the excellent book of E. Paradis. However I recently discovered the pPCA method (as introduced by Jombart et al. 2010) and i'm very interested in it to work on phylogenetic signal but I still have some questions... 1. I'm a long time user of ade4 to perform multivariate analysis. For a classic PCA I usually calculate the % of variance taking account by each axis using : myPCA$eig/sum(myPCA$eig) * 100 I'd just like to be sure I can do the same with a pPCA, using absolute values of the eigenvalues, e.g.: abs(myPPCA$eig)/sum(abs(myPPCA$eig)) * 10 2. In their paper, Jombart et al. present some figures where they sometimes exclude directly the local or the global principal component because they know it doesn't exist (these are simulated data). Is there a way to test the global vs the local component with real data? With my own data I have a very low local eigenvector so I wonder if I could only focus on global structure. Can I justify this choice with statistics? 3. I think it could be interesting to play with the species coordinates especially with the global component. But does it make sense to assess the phylogenetic signal or to estimate ancestral characters on these constrained data? I'm a little doubtful about that and your point of view is welcome. Thank you for your help. François KECK ___ R-sig-phylo mailing list - R-sig-phylo@r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-phylo Searchable archive at http://www.mail-archive.com/r-sig-phylo@r-project.org/
Re: [R-sig-phylo] Some questions about pPCA
Hello François, 1. In pPCA, the sum of the eigenvalues is often meaningless, because it can be a mixture of large positive and negative values. So this ratio is no longer relevant. Selection of eigenvalues can be based on the amount of variance and autocorrelation (Moran's I) represented (each eigenvalue is a product of the two). See summary.ppca and screeplot.ppca. 2. The best way is testing positive/negative phylogenetic autocorrelation (global/local structures in the paper's terminology) beforehand. See section 3.1 of the vignette Quantifying and testing phylogenetic signal - abouheif.moran will test all variables at once (just make sure to use the same measure of proximity in the pPCA). 3. As you suspected, testing phylogenetic signal of pPCA components is meaningless, as these synthetic variables are already optimized for phylogenetic signal. Estimating ancestral states is always possible; I can see at least two ways of doing it: a) reconstruct the ancestral state of every traits, and then compute the coordinates of the nodes on the pPCA axis using the loadings of the analysis. In this case, nodes are used as 'supplementary individuals'. b) reconstruct directly the principal component of pPCA; in this case, the component needs to have a clear-cut interpretation. Cheers Thibaut From: r-sig-phylo-boun...@r-project.org [r-sig-phylo-boun...@r-project.org] on behalf of Francois KECK [francois.k...@thonon.inra.fr] Sent: 11 February 2013 10:05 To: r-sig-phylo@r-project.org Subject: [R-sig-phylo] Some questions about pPCA Dear all, I'm a new subscriber to this list since I just started to play with phylogenetic data with R. The task is facilitated by reading the excellent book of E. Paradis. However I recently discovered the pPCA method (as introduced by Jombart et al. 2010) and i'm very interested in it to work on phylogenetic signal but I still have some questions... 1. I'm a long time user of ade4 to perform multivariate analysis. For a classic PCA I usually calculate the % of variance taking account by each axis using : myPCA$eig/sum(myPCA$eig) * 100 I'd just like to be sure I can do the same with a pPCA, using absolute values of the eigenvalues, e.g.: abs(myPPCA$eig)/sum(abs(myPPCA$eig)) * 10 2. In their paper, Jombart et al. present some figures where they sometimes exclude directly the local or the global principal component because they know it doesn't exist (these are simulated data). Is there a way to test the global vs the local component with real data? With my own data I have a very low local eigenvector so I wonder if I could only focus on global structure. Can I justify this choice with statistics? 3. I think it could be interesting to play with the species coordinates especially with the global component. But does it make sense to assess the phylogenetic signal or to estimate ancestral characters on these constrained data? I'm a little doubtful about that and your point of view is welcome. Thank you for your help. François KECK ___ R-sig-phylo mailing list - R-sig-phylo@r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-phylo Searchable archive at http://www.mail-archive.com/r-sig-phylo@r-project.org/ ___ R-sig-phylo mailing list - R-sig-phylo@r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-phylo Searchable archive at http://www.mail-archive.com/r-sig-phylo@r-project.org/
Re: [R-sig-phylo] Some questions about pPCA
Hello Thibaut, Thank you for these clarifications. About 2: I understand how to use the abouheif test to detect the phylogenetic signal (up to now I used it with the patristic matrix of proximities). But I don't know how to use it to test the anti-signal. Absence of signal is not anti-signal, or missed something? Cheers François Hello François, 1. In pPCA, the sum of the eigenvalues is often meaningless, because it can be a mixture of large positive and negative values. So this ratio is no longer relevant. Selection of eigenvalues can be based on the amount of variance and autocorrelation (Moran's I) represented (each eigenvalue is a product of the two). See summary.ppca and screeplot.ppca. 2. The best way is testing positive/negative phylogenetic autocorrelation (global/local structures in the paper's terminology) beforehand. See section 3.1 of the vignette Quantifying and testing phylogenetic signal - abouheif.moran will test all variables at once (just make sure to use the same measure of proximity in the pPCA). 3. As you suspected, testing phylogenetic signal of pPCA components is meaningless, as these synthetic variables are already optimized for phylogenetic signal. Estimating ancestral states is always possible; I can see at least two ways of doing it: a) reconstruct the ancestral state of every traits, and then compute the coordinates of the nodes on the pPCA axis using the loadings of the analysis. In this case, nodes are used as 'supplementary individuals'. b) reconstruct directly the principal component of pPCA; in this case, the component needs to have a clear-cut interpretation. Cheers Thibaut From: r-sig-phylo-boun...@r-project.org [r-sig-phylo-boun...@r-project.org] on behalf of Francois KECK [francois.k...@thonon.inra.fr] Sent: 11 February 2013 10:05 To: r-sig-phylo@r-project.org Subject: [R-sig-phylo] Some questions about pPCA Dear all, I'm a new subscriber to this list since I just started to play with phylogenetic data with R. The task is facilitated by reading the excellent book of E. Paradis. However I recently discovered the pPCA method (as introduced by Jombart et al. 2010) and i'm very interested in it to work on phylogenetic signal but I still have some questions... 1. I'm a long time user of ade4 to perform multivariate analysis. For a classic PCA I usually calculate the % of variance taking account by each axis using : myPCA$eig/sum(myPCA$eig) * 100 I'd just like to be sure I can do the same with a pPCA, using absolute values of the eigenvalues, e.g.: abs(myPPCA$eig)/sum(abs(myPPCA$eig)) * 10 2. In their paper, Jombart et al. present some figures where they sometimes exclude directly the local or the global principal component because they know it doesn't exist (these are simulated data). Is there a way to test the global vs the local component with real data? With my own data I have a very low local eigenvector so I wonder if I could only focus on global structure. Can I justify this choice with statistics? 3. I think it could be interesting to play with the species coordinates especially with the global component. But does it make sense to assess the phylogenetic signal or to estimate ancestral characters on these constrained data? I'm a little doubtful about that and your point of view is welcome. Thank you for your help. François KECK ___ R-sig-phylo mailing list - R-sig-phylo@r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-phylo Searchable archive at http://www.mail-archive.com/r-sig-phylo@r-project.org/ ___ R-sig-phylo mailing list - R-sig-phylo@r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-phylo Searchable archive at http://www.mail-archive.com/r-sig-phylo@r-project.org/
Re: [R-sig-phylo] Some questions about pPCA
Hello, I don't know what anti-signal is. Do you mean negative phylogenetic autocorrelation? When there is negative autocorrelation, Moran's index / Abouheif's statistic is smaller than permuted values (use 'plot' on the krandtest object). Cheers Thibaut From: r-sig-phylo-boun...@r-project.org [r-sig-phylo-boun...@r-project.org] on behalf of Francois KECK [francois.k...@thonon.inra.fr] Sent: 11 February 2013 13:47 To: r-sig-phylo@r-project.org Subject: Re: [R-sig-phylo] Some questions about pPCA Hello Thibaut, Thank you for these clarifications. About 2: I understand how to use the abouheif test to detect the phylogenetic signal (up to now I used it with the patristic matrix of proximities). But I don't know how to use it to test the anti-signal. Absence of signal is not anti-signal, or missed something? Cheers François Hello François, 1. In pPCA, the sum of the eigenvalues is often meaningless, because it can be a mixture of large positive and negative values. So this ratio is no longer relevant. Selection of eigenvalues can be based on the amount of variance and autocorrelation (Moran's I) represented (each eigenvalue is a product of the two). See summary.ppca and screeplot.ppca. 2. The best way is testing positive/negative phylogenetic autocorrelation (global/local structures in the paper's terminology) beforehand. See section 3.1 of the vignette Quantifying and testing phylogenetic signal - abouheif.moran will test all variables at once (just make sure to use the same measure of proximity in the pPCA). 3. As you suspected, testing phylogenetic signal of pPCA components is meaningless, as these synthetic variables are already optimized for phylogenetic signal. Estimating ancestral states is always possible; I can see at least two ways of doing it: a) reconstruct the ancestral state of every traits, and then compute the coordinates of the nodes on the pPCA axis using the loadings of the analysis. In this case, nodes are used as 'supplementary individuals'. b) reconstruct directly the principal component of pPCA; in this case, the component needs to have a clear-cut interpretation. Cheers Thibaut From: r-sig-phylo-boun...@r-project.org [r-sig-phylo-boun...@r-project.org] on behalf of Francois KECK [francois.k...@thonon.inra.fr] Sent: 11 February 2013 10:05 To: r-sig-phylo@r-project.org Subject: [R-sig-phylo] Some questions about pPCA Dear all, I'm a new subscriber to this list since I just started to play with phylogenetic data with R. The task is facilitated by reading the excellent book of E. Paradis. However I recently discovered the pPCA method (as introduced by Jombart et al. 2010) and i'm very interested in it to work on phylogenetic signal but I still have some questions... 1. I'm a long time user of ade4 to perform multivariate analysis. For a classic PCA I usually calculate the % of variance taking account by each axis using : myPCA$eig/sum(myPCA$eig) * 100 I'd just like to be sure I can do the same with a pPCA, using absolute values of the eigenvalues, e.g.: abs(myPPCA$eig)/sum(abs(myPPCA$eig)) * 10 2. In their paper, Jombart et al. present some figures where they sometimes exclude directly the local or the global principal component because they know it doesn't exist (these are simulated data). Is there a way to test the global vs the local component with real data? With my own data I have a very low local eigenvector so I wonder if I could only focus on global structure. Can I justify this choice with statistics? 3. I think it could be interesting to play with the species coordinates especially with the global component. But does it make sense to assess the phylogenetic signal or to estimate ancestral characters on these constrained data? I'm a little doubtful about that and your point of view is welcome. Thank you for your help. François KECK ___ R-sig-phylo mailing list - R-sig-phylo@r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-phylo Searchable archive at http://www.mail-archive.com/r-sig-phylo@r-project.org/ ___ R-sig-phylo mailing list - R-sig-phylo@r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-phylo Searchable archive at http://www.mail-archive.com/r-sig-phylo@r-project.org/ ___ R-sig-phylo mailing list - R-sig-phylo@r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-phylo Searchable archive at http://www.mail-archive.com/r-sig-phylo@r-project.org/
Re: [R-sig-phylo] Some questions about pPCA
Yes, I meant negative phylogenetic autocorrelation. With my data, permutations tests detect positive autocorrelation, no doubt on that. But there may exist overdispersion between some phylogenetically close species. So I would like to compare the global and local components. Maybe I can just say that there is 2.5 times more variability taking account by the global component. == pPCA eigenvalues decomposition == eig var moran Axis 1 119.8040 652.7595 0.1835347 Axis 4 -27.1536 238.8116 -0.1137030 Cheers François Hello, I don't know what anti-signal is. Do you mean negative phylogenetic autocorrelation? When there is negative autocorrelation, Moran's index / Abouheif's statistic is smaller than permuted values (use 'plot' on the krandtest object). Cheers Thibaut From: r-sig-phylo-boun...@r-project.org [r-sig-phylo-boun...@r-project.org] on behalf of Francois KECK [francois.k...@thonon.inra.fr] Sent: 11 February 2013 13:47 To: r-sig-phylo@r-project.org Subject: Re: [R-sig-phylo] Some questions about pPCA Hello Thibaut, Thank you for these clarifications. About 2: I understand how to use the abouheif test to detect the phylogenetic signal (up to now I used it with the patristic matrix of proximities). But I don't know how to use it to test the anti-signal. Absence of signal is not anti-signal, or missed something? Cheers François Hello François, 1. In pPCA, the sum of the eigenvalues is often meaningless, because it can be a mixture of large positive and negative values. So this ratio is no longer relevant. Selection of eigenvalues can be based on the amount of variance and autocorrelation (Moran's I) represented (each eigenvalue is a product of the two). See summary.ppca and screeplot.ppca. 2. The best way is testing positive/negative phylogenetic autocorrelation (global/local structures in the paper's terminology) beforehand. See section 3.1 of the vignette Quantifying and testing phylogenetic signal - abouheif.moran will test all variables at once (just make sure to use the same measure of proximity in the pPCA). 3. As you suspected, testing phylogenetic signal of pPCA components is meaningless, as these synthetic variables are already optimized for phylogenetic signal. Estimating ancestral states is always possible; I can see at least two ways of doing it: a) reconstruct the ancestral state of every traits, and then compute the coordinates of the nodes on the pPCA axis using the loadings of the analysis. In this case, nodes are used as 'supplementary individuals'. b) reconstruct directly the principal component of pPCA; in this case, the component needs to have a clear-cut interpretation. Cheers Thibaut From: r-sig-phylo-boun...@r-project.org [r-sig-phylo-boun...@r-project.org] on behalf of Francois KECK [francois.k...@thonon.inra.fr] Sent: 11 February 2013 10:05 To: r-sig-phylo@r-project.org Subject: [R-sig-phylo] Some questions about pPCA Dear all, I'm a new subscriber to this list since I just started to play with phylogenetic data with R. The task is facilitated by reading the excellent book of E. Paradis. However I recently discovered the pPCA method (as introduced by Jombart et al. 2010) and i'm very interested in it to work on phylogenetic signal but I still have some questions... 1. I'm a long time user of ade4 to perform multivariate analysis. For a classic PCA I usually calculate the % of variance taking account by each axis using : myPCA$eig/sum(myPCA$eig) * 100 I'd just like to be sure I can do the same with a pPCA, using absolute values of the eigenvalues, e.g.: abs(myPPCA$eig)/sum(abs(myPPCA$eig)) * 10 2. In their paper, Jombart et al. present some figures where they sometimes exclude directly the local or the global principal component because they know it doesn't exist (these are simulated data). Is there a way to test the global vs the local component with real data? With my own data I have a very low local eigenvector so I wonder if I could only focus on global structure. Can I justify this choice with statistics? 3. I think it could be interesting to play with the species coordinates especially with the global component. But does it make sense to assess the phylogenetic signal or to estimate ancestral characters on these constrained data? I'm a little doubtful about that and your point of view is welcome. Thank you for your help. François KECK ___ R-sig-phylo mailing list - R-sig-phylo@r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-phylo Searchable archive at http://www.mail-archive.com/r-sig-phylo@r-project.org/ ___ R-sig-phylo mailing list - R-sig-phylo@r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-phylo Searchable archive at
Re: [R-sig-phylo] Some questions about pPCA - phylogenetic anti-signal
You can test for anti-signal using the randomization procedures of: Blomberg, S. P., T. Garland, Jr., and A. R. Ives. 2003. Testing for phylogenetic signal in comparative data: behavioral traits are more labile. Evolution 57:717-745. See page 719. Cheers, Ted Theodore Garland, Jr., Professor Department of Biology University of California, Riverside Riverside, CA 92521 Office Phone: (951) 827-3524 Facsimile: (951) 827-4286 = Dept. office (not confidential) Email: tgarl...@ucr.edu http://www.biology.ucr.edu/people/faculty/Garland.html http://scholar.google.com/citations?hl=enuser=iSSbrhwJ Inquiry-based Middle School Lesson Plan: Born to Run: Artificial Selection Lab http://www.indiana.edu/~ensiweb/lessons/BornToRun.html From: r-sig-phylo-boun...@r-project.org [r-sig-phylo-boun...@r-project.org] on behalf of Francois KECK [francois.k...@thonon.inra.fr] Sent: Monday, February 11, 2013 5:47 AM To: r-sig-phylo@r-project.org Subject: Re: [R-sig-phylo] Some questions about pPCA Hello Thibaut, Thank you for these clarifications. About 2: I understand how to use the abouheif test to detect the phylogenetic signal (up to now I used it with the patristic matrix of proximities). But I don't know how to use it to test the anti-signal. Absence of signal is not anti-signal, or missed something? Cheers François Hello François, 1. In pPCA, the sum of the eigenvalues is often meaningless, because it can be a mixture of large positive and negative values. So this ratio is no longer relevant. Selection of eigenvalues can be based on the amount of variance and autocorrelation (Moran's I) represented (each eigenvalue is a product of the two). See summary.ppca and screeplot.ppca. 2. The best way is testing positive/negative phylogenetic autocorrelation (global/local structures in the paper's terminology) beforehand. See section 3.1 of the vignette Quantifying and testing phylogenetic signal - abouheif.moran will test all variables at once (just make sure to use the same measure of proximity in the pPCA). 3. As you suspected, testing phylogenetic signal of pPCA components is meaningless, as these synthetic variables are already optimized for phylogenetic signal. Estimating ancestral states is always possible; I can see at least two ways of doing it: a) reconstruct the ancestral state of every traits, and then compute the coordinates of the nodes on the pPCA axis using the loadings of the analysis. In this case, nodes are used as 'supplementary individuals'. b) reconstruct directly the principal component of pPCA; in this case, the component needs to have a clear-cut interpretation. Cheers Thibaut From: r-sig-phylo-boun...@r-project.org [r-sig-phylo-boun...@r-project.org] on behalf of Francois KECK [francois.k...@thonon.inra.fr] Sent: 11 February 2013 10:05 To: r-sig-phylo@r-project.org Subject: [R-sig-phylo] Some questions about pPCA Dear all, I'm a new subscriber to this list since I just started to play with phylogenetic data with R. The task is facilitated by reading the excellent book of E. Paradis. However I recently discovered the pPCA method (as introduced by Jombart et al. 2010) and i'm very interested in it to work on phylogenetic signal but I still have some questions... 1. I'm a long time user of ade4 to perform multivariate analysis. For a classic PCA I usually calculate the % of variance taking account by each axis using : myPCA$eig/sum(myPCA$eig) * 100 I'd just like to be sure I can do the same with a pPCA, using absolute values of the eigenvalues, e.g.: abs(myPPCA$eig)/sum(abs(myPPCA$eig)) * 10 2. In their paper, Jombart et al. present some figures where they sometimes exclude directly the local or the global principal component because they know it doesn't exist (these are simulated data). Is there a way to test the global vs the local component with real data? With my own data I have a very low local eigenvector so I wonder if I could only focus on global structure. Can I justify this choice with statistics? 3. I think it could be interesting to play with the species coordinates especially with the global component. But does it make sense to assess the phylogenetic signal or to estimate ancestral characters on these constrained data? I'm a little doubtful about that and your point of view is welcome. Thank you for your help. François KECK ___ R-sig-phylo mailing list - R-sig-phylo@r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-phylo Searchable archive at http://www.mail-archive.com/r-sig-phylo@r-project.org/ ___ R-sig-phylo mailing list - R-sig-phylo@r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-phylo Searchable archive at http://www.mail-archive.com/r-sig-phylo@r-project.org/