Thorne Kishino Felsenstein

From Biowiki
Jump to: navigation, search

The TKF91 and TKF92 models


The Thorne-Kishino-Felsenstein 1991 model (TKF91) is the first explicit derivation of a pairwise sequence alignment algorithm from a continuous-time Markov chain defined on whole sequences. It is a landmark paper in Statistical Alignment.

Alignment algorithms for the TKF91 model are implemented in the Handel Package.

Empirical drawbacks of the TKF91 model revolve around its simplicity, in particular the assumption that each indel event involves only a single residue. This is the equivalent of the "linear gap penalty" in bioinformatics, and gives unrealistic (gappy) alignments.


The followup "TKF92 model" attempts to introduce an "affine gap penalty" by breaking the sequence into indivisible fragments. Each indel event involves exactly one fragment; indel events never overlap, and the math turns out similar to the TKF91 model, but with "residues" replaced by "fragments".

TKF92 can be viewed as an approximation to a Long Indel model.

For more context, see the Phylogenetic Alignment Reader.

-- Ian Holmes - 23 Apr 2008