Tuesday, March 25, 2008

24 March 2008

It has been a while since I posted (or we met!) but a new year is underway and we should continue on. We have had a visitor to the site who had the following question:

Emily: I have a question about the gap handling. How can we deal with the gaps in ML or NJ methods. In the blog you said that "Only in Parsimony methods can this be automatically done (as a 5th or 21st option for nucleotide and amino acids respectively). "?

Nadir was good enough to offer a response:

Hi, Dealing with gaps in distance methods is possible but implies that you do not consider any model of evolution, but simply compute a similarity/distance matrix. Jaccard distances are fine to build a distance matrix (you might further use neighbor.exe [Phylip package] to construct your NJ tree). Regarding ML, Felsenstein has developed restML, a module within the Phylip package, able to deal with restriction fragments.

Since a gap can be seen - theoretically - as a potentially polymorphic site for restriction enzymes, it is possible to run restML on a matrix of presence/absence of gaps. However, Felsenstein's algorithm assumes a fixed length of the restriction site (eg, 5 nucleotides), which is very unlikely to be the case when considering different gap regions. If you want to perform analyses on gaps using a probabilistic algorithm, I would rather suggest to consider the use of MrBayes (you can define a gap matrix as being of the type datatype=restriction).









No comments: