Calcium-Binding Protein Protocols Calcium-Binding Protein Protocols

240 Weljie and Heringa
lineages. The method of Saitou and Nei (37) (called NJ), is acclaimed by many
workers in phylogenetic analysis. The method relies on a protocol of progressive
pairwise joining of nearest sequences (each sequence being represented by a
node) such that each time two nodes are joined, they are represented by an internal
node. The two nodes selected at each step for joining are those that keep the
overall tree length at a minimum. The NJ method has the advantage over the
UPGMA technique in that it does not use the evolutionary (dis)similarity among
groups, but merely is a strategy to join sequences and calculate branch lengths
without assumptions regarding the rate of evolution. A general advantage of distance
methods over parsimony and maximum likelihood methods is that they
usually are much less CPU intensive as they employ a fixed strategy to arrive at a
final tree without the need to sample the complete and vast tree space.
The simplest way to calculate the distance between sequences is by using
the percent divergence, which for two aligned sequences involves counting
the number of nonidentical matches (ignoring positions with gaps) divided
by the number of positions considered. When a multiple alignment is used,
all positions containing a gap in any of the sequences are commonly ignored.
The real evolutionary time between the divergence of two sequences depends
on the speed of the evolutionary clock, a matter of ongoing controversy. Even
under a uniform clock, sequence identity as a measure of distance underestimates
the real number of mutations. Certainly in diverged sequences there is
an increasing chance that multiple substitutions have occurred at a site. The
greater the divergence, the more the evolutionary times are underestimated.
Kimura (59) corrected for this effect by curve fitting such that the corrected
evolutionary time from the distance K (percent divergence divided by 100) is
given by corrected K = –ln(1.0 – K – K 2 / 5.0). The formula applies to cases
with a reasonably uniform evolutionary clock and sequence identies from 15%
and fits the data well from identities higher than 35%. It is good practice to
start tree-building routines from such corrected sequence distances, which
can be done by the PHYLIP package (60–62).
Because evolutionary trees may be a result of local traps in the search space, it
is important to estimate the significance of a particular tree topology and associated
branch lengths. Felsenstein (63) introduced the concept of bootstrapping,
which involves resampling of the data such that the alignment positions are randomly
selected and placed in some order and then a tree is generated by the
original method. This process is repeated a statistically significant number of
times. Comparison of frequencies at which the N-3 internal branches (N is the
number of sequences) occur in the original tree and those from bootstrapping
allow probability estimates of significance. It is common practice to consider
frequencies higher than or equal to 95% as supportive for the occurrence of an
original tree branch in the bootstrapped trees. In this way, bootstrapping tests the
stability of groupings given the data set and the method, thus lessening the chance
of incorrect tree structures caused by conservative and/or back mutations.
Bootstrapping can be performed for any method that generates a tree from a mul-