From: ncbi-admin@ncbi.nlm.nih.gov on behalf of anna panchenko
[panch@ncbi.nlm.nih.gov]
Sent: Monday, March 05, 2001 3:36 PM
To: ncbi-seminar@karabas.nlm.nih.gov
Subject: NCBI seminar, 8-th floor conference room, Bldg 38A!


NCBI seminar, on the 6-th of March, at 11 a.m. in the 8-th floor 
conference room in Bldg 38A.

 
           FINDING AN OPTIMAL STRATEGY FOR CONSTRUCTING THE MINIMAL SET
     OF PROFILES REQUIRED TO RECOGNIZE MEMBERS OF A DIVERSE PROTEIN FAMILY.
 
                             Anna R. Panchenko.
 
        Sequence comparison methods based on position specific score
matrices, or profiles, have proven useful tools for recognition of the
divergent members of protein families. In our previous work, we found that
profile recognition sensitivity depends on the diversity of the sequences
included in the alignment, with an optimum around 30-50% average pairwise
identity. 
        Below this range the sensitivity of profiles decreases, which
suggests a strategy for constructing the minimal set of PSSM/profiles: one
should divide a diverse family into subfamilies in order to recognize remote
family members. We tested this strategy using the SMART collection of domains
as a starting point. I am going to discuss the factors, which influence the
success of this strategy to increase the recognition sensitivity. 
        At the same time, if a profile is constructed from the alignment of
closely related sequences, the search procedure will soon reach the boundary
of detectable similarities without expanding far beyond the starting point. 
In this case two or more profiles should be merged together to form a profile
informative enough to recognize distant family members. In order to do this,
we developed a Monte Carlo based profile-profile alignment method, which can
be successfully used for profile-profile alignments and fold recognition as
well.