Lawren Smithline

13 11 03

Here  is a writeup of the method applied below to Mus Musculus zfp111 and zfp235.

06 10 03

Here  is the (developing) story of comparing Mus Musculus zfp111 and zfp235.

26 03 03

Here  is an example of another derivative of the TKF
model showing a pair of sequences with multiple regions
of good alignment, in different permutations.
 

08 03 03

Below is a representation of possible alignments of 3.5 hectobase segments from two related genomes. The data came from the Whitehead Institute.  The new method is a derivative of the TKF Markov  model (minus most of the reestimation bells and whistles). The method reports aligments as  a distribution, rather than a single best result.
The deepest red is the most probable.

Prob(alignment passes through a particular purple point)
is less than
exp(-50).Prob(alignment passes through a particular red point).

Here's another picture showing the same two sequences aligned with a related model.

The differences are analogous to cancelling the gap penalty for terminal gaps. This is expressed by running the TKF model as the sequence extension process instead of the insertion process along the trailing edges of the array.  This picture shows that this modified TKF process  prefers a terminal gap. The other modification to the this process was the assumption that the sequences are selections from essentially infinite strings. This causes a bias away from shortest alignment, or removes a bias towards substitution events.