as by the N- or C-terminus of the protein or by an adjacent well-characterized domain.
Thus, not all local similarities claimed to be ‘homology domains’ really are domains
in the structural sense, and even if they are, the position of the boundaries can deviate.
Notwithstanding those caveats, most of the proposed homology domains that occur
in diverse sets of proteins correspond nicely to domains in the structural sense. Since
most domains, in their cut-and-paste process, preserve not only their structure but
also their function, it is frequently possible to attach functional labels to particular
domain types. In extreme cases, the property of a novel protein can be predicted from
the functional domains contained in its sequence.
The term ‘homology domain’ or ‘functional domain’ should be used only for those
protein regions either that are known to be domains in the structural sense, or that
are at least predicted to fulfill that criterion. Conserved sequence regions that are too
short to fold independently from the rest of the protein should rather be referred to
as ‘motifs’. Conserved motifs can have important roles, too: they are frequently
recognized specifically by other proteins or domains.
1.2
Domain detection methodsAs mentioned in the previous paragraph, homology domains can be detected by
sequence analysis, where they appear as regions of detectable similarity embedded
into sequences that are otherwise unrelated. Thus, any tool for local alignment, such
as those using the Smith and Waterman algorithm (Smith and Waterman, 1981), is
suitable for detecting domains if they are moderately well conserved. It is a widely
held tenet that in distantly related proteins, the structure is much better conserved
than the sequence. A similar observation can be made for a protein’s function, some
aspects of which are frequently maintained at evolutionary distances where the
sequences no longer look similar. As a consequence, it can be expected that domains
exist, both in the structural and functional sense, which cannot be spotted easily by
sequence comparison alone. Although several ‘structural genomics’ projects have
gained momentum, it is still not an option to wait for all relevant structures to be
determined in order to identify a domain.
Recent years have seen not only a dramatic increase in the available sequence
material, but also a constant improvement in the available sequence analysis
techniques. In particular, the ‘sequence profile’ method (Gribskov et al., 1987) with
its recent extension to ‘generalized profiles’ (Bucher et al., 1996), and various ‘hidden
Markov model’ methods (HMM) (Baldi et al., 1994; Krogh et al., 1994; Eddy, 1998)
have proved very useful for detecting very weak sequence similarities. Some of their
properties make these two methods very well suited for domain detection purposes
(Hofmann, 2000). A significant number of homology domains have been identified
by either method, including most of the domains discussed in the following
paragraphs.
The first-time identification of a new homology domain typically leads to its
description by a generalized profile (GP) or an HMM. The domain descriptors are
FUNCTIONAL DOMAINS IN APOPTOSIS PROTEINS 73