The methods of the previous chapters were applied in order to cluster
all protein segments of length 50 in the SWISSPROT database.  To
summarize the computational procedure, we start by partitioning all
protein sequences longer than 50 amino acids in the SWISSPROT database
(release 30) into overlapping segments of 50 amino acids. This
partitioning yields a total of 543,627 segments.  A metric derived
from the Smith-Waterman dynamic programming measure of similarity
turns the space of protein segments into a finite metric space (see
section 4.1).  This space of segments is initially
embedded into Euclidean space of low dimension and small distortion
(see chapter 4).  A hierarchical clustering algorithm
is then applied to the embedded space with Euclidean distances.  A key
aspect of this stage is that the validity of our clustering is closely
monitored by cross validation (see chapter 5).
  
The processing of such large quantities of data requires extensive
computer resources. The embedding phase requires massive pairwise
calculations, and was made possible by using Compugen's Bioccelerator
that performs the SW dynamic programming algorithm.
The clustering phase is based on a computationally intensive protocol
that was made feasible by programming a parallel application to run on
the MOSIX distributed system.

The whole computational process was fully automatic, without any human
intervention or biological consideration.  On termination, when the
cross validation criteria allowed no further splitting, the process
yielded a tree of 106 clusters.  This chapter discusses the results of
this analysis.