From: Kann, Maricel (NIH/NLM/NCBI) [F]
Sent: Saturday, February 18, 2006 7:31 PM
To: NLM/NCBI List ncbi-seminar
Subject: CBB Seminar 11:00 Tue Feb 21 in B2 Library

GLOBAL (GLObal Blocks Aligned Locally)
A General Algorithm (with p-value) for Sequence Classification

The Needleman-Wunsch (1970) algorithm for global alignment could well be
considered the first bioinformatics paper.  Despite the obvious
importance of global alignment methods, in bioinformatics it is common
knowledge that there is no global alignment method with an associated
p-value.

This talk presents GLOBAL, a general method for combining local
alignments into a global alignment.  Because GLOBAL is based on local
alignments, it is amenable to BLAST heuristics in database searches,
which many other methods (e.g., hidden Markov models) are not.  The
present version of GLOBAL is tailored specifically for use in the CD
database, which models protein domains as blocks, separated by gaps of
variable length.  The blocks represent conserved secondary sequence
elements in domains; they are separated by gap regions, which are more
mutable and sometimes contain deletions.

Last week, Dr. John Spouge presented the basis of the GLOBAL p-value calculation along with
simulations indicating the accuracy of the p-value. In this seminar, I will present results of the evaluation 
of GLOBAL against rpsBLAST (reverse-position specific BLAST) and HMMer (Hidden Markov Models).
Our results suggest retrieval efficacies are in the following order: rpsBLAST < GLOBAL <
HMMer.  
 
 
Maricel Kann.
Postdoctoral Fellow
Email: kann@mail.nih.gov
Phone: +1 (301) 402-3010