CBB seminar
Mon, 11 Dec 2000, 11:00
8th floor conference room, Bldg 38A

Protein Sequence Clustering.

Yuri Wolf.

Two approaches to protein sequence clustering, focusing on different 
neighboring criteria, will be discussed. The quantitative results of 
lineage-specific family clustering of eukaryotic proteomes will be 
presented.


1. Clustering with fixed thresholds (BLASTCLUST). Defines neighboring 
relationships via thresholds on similarity degree and extent, then 
performs single-linkage clustering. Used to make "non-redundant" 
sequence sets and to obtain wide clusters of similarly organized 
proteins.

2. Clustering of lineage-specific families. Consists of two stages. On 
the first (accretion) stage sequences belonging to a given lineage are 
clustered by a single-linkage algorithm on the basis of comparison of 
similarity within the taxonomic lineage to that outside the lineage. On 
the second (refinement) stage best "alien" hits are added to the 
original clusters and "native"-only subclusters are found using UPGMA 
method. Used to obtain sets of lineage-specific paralogs for 
evolutionary studies.