To validate this result further, Husník and colleagues [ 2] applied another approach, the removal of sites.
As we have discussed below, our analyses also suggest that removal of sites rather than individual genes provides a better strategy for dealing with this problem.
If the distribution of missing data is not random (an important concept in statistics in general, Allison 2002), the removal of sites with missing data could bias tree inference.
It is important to note that the slowest evolving positions can be misleading particularly with excessive removal of sites, as the number of characters for reconstruction will decrease with every cycle, therefore caution must be taken in applying this method.
In general, for phylogenetic data, it seems unlikely that data will be missing at random, and therefore, the decision regarding the inclusion or removal of sites with missing data may be very influential.
The removal of sites with poor frequency estimates (| F E | > 0.1 in at least two populations) results in a marked change in the distribution of H, with a significant drop in the frequency of sites with large H and a shift in the distribution toward lower values.
With the removal of sites with heterogeneous profile affiliations across clades, support for the monophyly of Eumetazoa (top of Figure 5) increases steadily, up to about one, and decreases to 0.6 in the smallest dataset, probably due to the limited amount of data.
If the substrate protein contains many sites, removal of one site (e.g. due to adverse mutations) is less likely to cause its function to break down, i.e. the system can tolerate better perturbations.
If the site by gender interaction effect was retained, this indicated that gender effects were inconsistent across sites, while removal of the site by gender interaction effect indicated that any apparent inconsistencies across sites were merely due to sampling variation.
Screen colonies for removal of NotI site and addition of FseI site.
Removal of NotI site in the multiple cloning site of pRS314.
