Phylogeny and molecular evolution of green algae - Phycology ...

14 CHAPTER 1
Missing data
Deep phylogenies require the simultaneous analysis of many characters and many taxa (Delsuc et al.
2005). Individual, orthologous genes can be combined into a supermatrix which inevitably involves a
certain amount of missing data. Many studies have studied the effects of missing data on
phylogenetic reconstruction. A simulation study suggests that the placement of individual taxa in a
tree is robust to large amounts of missing data in the sequences of the taxa in question (up to 50%
under the simulated conditions) and that model-based methods can deal with even greater amounts
of missing data (Wiens 2005). Another simulations study demonstrates that Bayesian analyses are
even more robust to missing data, i.e. the phylogenetic position of taxa with 95% of missing data in
their sequence is still accurate, as long as the total number of characters in the dataset is large
(Wiens and Moen 2008). Studies of empirical datasets have shown that datasets with up to 92% of
missing data are still able to provide insights into various parts of the tree of life (Driskell et al. 2004,
Philippe et al. 2004, Delsuc et al. 2005).
Models of sequence evolution
The General Time Reversible (GTR) model and its simpler variants include one or more parameters to
describe the substitution rate between the different bases. The GTR model uses a set of parameters
to describe the relative substitution rate between all combinations of bases (AC, AG, AT, CG, CT, and
GT). The simpler models only consider transitions versus transversions or attribute an equal
substitution rate to all possible changes. A second important component of a model are the base
frequencies. They can be calculated directly from the dataset (‘empirical’ base frequencies) or
optimized along with the other parameters of the model. A third common element of the model
allows for variations of evolutionary rate across site (e.g. different codon positions in protein coding
genes, loops and stems in ribosomal DNA). Such among site rate variation is commonly accounted for
by assuming that the site rates follow a gamma distribution and/or by incorporating a proportion of
invariable sites.
Partitioning strategies
A supermatrix, a dataset composed of different genes, often demands data partitioning to account
for across site heterogeneity in evolutionary rate (Delsuc et al. 2005). Therefore, careful attention
has to be paid to the selection of suitable partitioning strategies (Brown and Lemmon 2007, Li et al.
2008, Verbruggen and Theriot 2008). Protein coding genes usually benefit from partitioning into
codon position. Empirical studies showed that codon position models perform better than models
which do not take codon position into account (Shapiro et al. 2006). In order to accommodate
differences in evolutionary rate among partitions rate multipliers can be used.