Different genes evolve at different rates.  It is possible to build trees from a mix of data such as you are describing, but the results need to be understood in the context of the data used.  With the explosion of data available in databases today, it is quite common to see researchers gathering large but sparsely populated data matrices from which to build phylogenies.    The resulting trees quite often have some species on long branches because only rapidly evolving genes (mitochondrial genes for example) were available from those species while other species are on short branches because only more conserved genes were available, and I have seen authors claim that this indicates different rates of evolution of the species as a whole when in fact it does not.

Most phylogenetic methods require that there be at least some overlap of comparable sites across all species in the set.  And it may be more instructive to build a tree from each gene and then compare the trees, than to attempt to build one supertree when the  larger set is a sparsely populated matrix.

It is a little difficult to give advice in general about such things, because not all types of organisms have similar evolution.  For example some species have more geographical or other barriers preventing gene flow between populations as they evolve.  Are you dealing with viruses, plants, fungi, bacteria, vertebrates, insects?

One example I am familiar with, which has their complete data matrix available in TreeBase so you can explore it, is a study of avian evolution by Hackett et al 2008:  http://www.sciencemag.org/content/320/5884/1763    http://treebase.org/treebase-web/search/study/summary.html?id=11811

Different genes evolve at different rates.  It is possible to build trees from a mix of data such as you are describing, but the results need to be understood in the context of the data used.  With the explosion of data available in databases today, it is quite common to see researchers gathering large but sparsely populated data matrices from which to build phylogenies.    The resulting trees quite often have some species on long branches because only rapidly evolving genes (mitochondrial genes for example) were available from those species while other species are on short branches because only more conserved genes were available, and I have seen authors claim that this indicates different rates of evolution of the species as a whole when in fact it does not.

Most phylogenetic methods require that there be at least some overlap of comparable sites across all species in the set.  And it may be more instructive to build a tree from each gene and then compare the trees, than to attempt to build one supertree when the  larger set is a sparsely populated matrix.

It is a little difficult to give advice in general about such things, because not all types of organisms have similar evolution.  For example some species have more geographical or other barriers preventing gene flow between populations as they evolve.  Are you dealing with viruses, plants, fungi, bacteria, vertebrates, insects?

One example I am familiar with, which has their complete data matrix available in TreeBase so you can explore it, is a study of avian evolution by Hackett et al 2008:  http://www.sciencemag.org/content/320/5884/1763    http://treebase.org/treebase-web/search/study/summary.html?id=11811

Hi Peter,

The concerns raised by Dr. Brian are really valid not only if the data is missing but also in the regular molecular phylogenetic analysis, for the reasons also explained above. I also have some more points:

1) First of all for such analysis use genes/regions which are present is all the taxa to be analyzed...since the analysis investigates ancestral realtionships...in the absence of a region in a taxa there is no information we can get.

2) Analysis can be done in a variety of ways and using multiple algorithms...no one method is perfect...based on the type of species, rate of evolution of the genomic region one need to assess relationships by more than one method

3) genes with missing regions/ partially covered or actual deletions can also be analyzed by several ways. One can remove the problematic regions (indels/partially covered regions) during analysis where these are removed from all the sequences  or may  remove them in a pariwise manner during comparison

4) You may  combine multiple alleles or single allele of samples in the same analysis ...this will be informative regarding the divergence and evolution and distribution of alleles in the taxa.

5) regarding multiple sequences in the same genome (or paralogs)...these cane be also included in the same analysis..and based on the preliminary results can be included or removed for refined analysis later. Several papers are available on such analysis in ITS and other regions...including multiple (even intragenomic sequences) will be more informative for such analysis.

6) Use more than one algorithm for analysis and do include statistical analysis of the data for significance..

best regards

Hello Peter,

Here is a recent paper you may find useful.

https://www.researchgate.net/publication/284273579_The_impact_of_missing_data_on_species_tree_estimation

Best Regards

Article The Impact of Missing Data on Species Tree Estimation

Hi Peter

My recommendation is to use a multispecies coalescent (MSC) approach, by which the species tree and gene trees are estimated simultaneously. The different sequences, including missing ones, have to be a priori assigned to species, and thus more than one allele per species can be entered and not all have to be from the same specimens.

This approach is far more accurate than concatenation (which should always be avoided if possible). The problem mentioned by others above concerning genes evolving at different rates is explicitly accounted for by a different rate being assigned to each gene/region.

I have used the *BEAST function in the BEAST software for this. However, with 12 genes, 50 species and some missing data, the analysis will certainly take a long time to converge and complete unless you have access to a CPU cluster.

Perhaps an acceptable compromise is to concatenate the sequences and analyse them in BEAST (not *BEAST!), under the assumptions that the gene trees are linked (i.e. each gene tree = 'concatenated tree' = species tree), the clocks (one for each gene) are linked, but that the rates (one for each gene) are unlinked. This would circumvent the computational problem in *BEAST, at least to some degree, but you would only get to include one allele per gene per species (or individual). The latter would at least be my choice for an initial, exploratory analysis.

Good luck with the analysis. Feel free to ask if you need assistance setting up any of the analyses.

Thomas

Apart from all the above stated, be sure to have at least one gene  to act as backbone. ;)

Interesting topic! And very informative to see what you're doing in molecular phylogeny to cope with it.

Missing data is a big issue in paleontology, where we only have morphological data to use for phylogenetic studies. However, there are several simulations and other studies which showed that this does not have to be a problem a priori. In fact, it seems that including characters or OTUs with a lot of missing data still improves the outcome of the phylogenetic tree, because it still adds information.

Below some of the studies mentioned:

Norell, M. A., and W. C. Wheeler. 2003. Missing entry replacement data analysis: a replacement approach to dealing with missing data in paleontological and total evidence data sets. Journal of Vertebrate Paleontology 23:275–283.

Pol, D., and I. H. Escapa. 2009. Unstable taxa in cladistic analysis: identification and the assessment of relevant characters. Cladistics 25:515–527.

Wiens, J. J. 1998. Does adding characters with missing data increase or decrease phylogenetic accuracy? Systematic Biology 47:625–640.

Wiens, J. J. 2005. Can incomplete taxa rescue phylogenetic analyses from long-branch attraction? Systematic Biology 54:731–742.

Wiens, J. J., and J. Tiu. 2012. Highly incomplete taxa can rescue phylogenetic analyses from the negative impacts of limited taxon sampling. PLoS ONE 7:e42925.

Wilkinson, M. 1995. Coping with abundant missing entries in phylogenetic inference using parsimony. Systematic Biology 44:501–514.

Wilkinson, M. 2003. Missing entries and multiple trees: instability, relationships, and support in parsimony analysis. Journal of Vertebrate Paleontology 23:311–323.