Domains classification methods

We have recently undertaken a reconsideration of cellulase sequences by analyzing each domain of the sequence separately. Here we describe initial observations and methodologies we are developing for the analysis and comparison of cellulase domains. Novel methods of classification will give greater insight into theoretical, evolutionary and practical approaches to the analysis of cellulase. 

This two-stage method, implemented in DALI, has been used to compare representatives from all nonhomologous (in sequence) families in the protein data bank. " More details are given in the DALI Domain Classification below. 

All current structural classification methods are based on the same scheme Protein structures are first divided into discrete, globular domains, which are then classified at the levels of (1) class, (2) folds, (3) superfa-milies, and (4) families. The differences among existing schemes come from the methods that define the domains and the procedures that classify. After reviewing the terms that define a classification, the three main protein structure classifications available, SCOP, CATH, and the DALI Domain Dictionary (DDD), will be described. Links to these databases and related services are listed in Table 7. 

SCOP, CATH, and DDD agree on most of their classifications, despite differences in the classification methods they have implemented, and in the rules of protein structure and taxonomy they are based on. Hadley and Jones were the first to publish a detailed comparison of the fold classifications produced by SCOP, CATH, and FSSP. They showed that the three classification systems tend to agree in most cases, and that the discrepancies and inconsistencies are accounted for by a small number of problems. Among these, the domain assignment plays a crucial role. As mentioned, the separation of proteins into domains is a difficult and often subjective process. Many... 

A priori it cannot be assumed that the relationships between spectral data and structural data are linear. A lot of effort therefore has been spent to improve spectra classification by the implementation of non-linear methods, especially by neural networks. Typical results show that neural networks are sometimes better than linear discrimination methods but not considerably better. Obviously the representation of spectra by appropriate features (which is a domain of chemistry) is more important than the applied classification methods. 

As a starting point in the description of the solid intermetallic phases it is useful to recall that their identification and classification requires information about their chemical composition and structure. To be consistent with other fields of descriptive chemistry, this information should be included in specific chemical and structural formulae built up according to well-defined rules. This task, however, in the specific domain of the intermetallic phases, or more generally in the area of solid-state chemistry, is much more complicated than for other chemical compounds. This complexity is related both to the chemical characteristics (formation of variable composition phases) and to the structural properties, since the intermetallic compounds are generally non-molecular in nature, while the conventional chemical symbolism has been mainly developed for the representation of molecular units. As a consequence there is no complete, or generally accepted, method of representing the formulae of intermetallic compounds. 

Selected entries from Methods in Enzymology [vol, page(s)] General Protein kinase classification, 200, 3 protein kinase catalytic domain sequence database identification of conserved features of primary structure and classification of family members,... 

While it is to some extent arbitrary, a classification of this kind provides a means of discussing some of the general features now emerging from studies of metallic oxides. We have stressed the evidence that a nonstoichiometric phase is disordered, but may be related to chemically similar phases of fixed composition where an anomaly of structure is ordered and identifiable by x-ray diffraction methods. Where such ordered phases are found, it is possible that features of them are retained as blocks or domains with short range order in the related berthollide. Efforts should be directed towards order-disorder effects, with a view to reconsidering the status of the nonstoichiometric compound with a very wide composition range. 

Recently, the Rosetta stone method has been introduced. This method uses over 20 completely sequenced genomes and analyzes evolutionary correlations of two domains being fused into one protein in one species and occurring in separate proteins in another species. From these classifications the method establishes pairwise links between functionally related proteins [94] and elicits putative protein-protein interactions [95]. 

A comprehensive description of the databases and methods for domain, family, and pattern identification is available in chapter 2. Therefore, in this chapter, the discussion of the application of family and domain information to function assignment will be limited to Pfam, the virtual standard in protein domain/family classification, and to InterPro and CDD, two resources that integrate multiple domain databases. In addition, I discuss tools that can be used to scan those databases, namely HMMER,... 

---