We invite you to the first biostatistics seminar in the 2008/9 series.
Tag Single Nucleotide Polymorphism selection based on clustering with
dominant sets and replicator dynamics
Florian Frommlet
Department of Statistics,
University of Vienna
Friday, 19th September
2pm
A2002
The last couple of years have seen an increasing number of algorithms on
tag Single Nucleotide Polymorphism (SNP) selection, where one is
searching for a minimal number of most informative SNPs. Many of these
algorithms are based on pairwise linkage disequilibrium (LD) patterns.
The set of SNPs together with a measure of pairwise LD (e.g. r2) can be
represented as an edge weighted graph. Pavan and Pelillo (2007) have
recently introduced the notion of dominant sets for edge weighted
graphs, which generalizes the concept of maximal cliques for ordinary
graphs. With ordinary graphs there is an intimate relationship between
the maximum clique problem and results from evolutionary game theory,
which allow us to compute maximal cliques via so called replicator
dynamics, a set of discrete time dynamical systems. This idea has now
been extended to edge weighted graphs, and replicator dynamics can be
used to compute dominant sets. In our case these are none other than
clusters of SNPs with large pairwise LD, such that LD between SNPs of
different clusters is minimal. Replicator dynamics then allows us to
choose an optimal representative of each cluster, which can be used as
a tag SNP. We discuss certain advantages of the new approach and compare
its performance with algorithms from the literature.
|