Reliability of score based alignments
vendredi 13 novembre 2009, 9h30 - 10h45
The search of sequence homology remains undoubtedly one of the most
critical problem in bioinformatics. Recently, a great deal of
attention has been drawn on probabilistic
alignments in comparison to the classical score-based ones.
Unfortunately, this probabilistic approach usually requires to use
pair hidden Markov models whose parameters cannot be easily connected to score-based
ones. I will present an alternative model (Miyazawa, 1995) which allows
for directly translating scoring functions into a probabilistic framework.
Using the forward and backward algorithm it is possible to compute the
marginalized posterior probabilities for a given alignment as a measure of
the reliability. We developed further algorithms that compute
exactly the posterior probabilities of the currencies of patterns responsible for so called gap biases.