Comparative modeling usually starts by searching the Protein Data Bank (PDB) of known protein structures using the target sequence as the query. This search is generally done by comparing the target sequence with the sequence of each of the structures in the database. A variety of sequence-sequence comparison methods can be used [25,26,27,28,29]. Frequently, availability of many sequences related to the target or potential templates allows more sensitive searching with sequence profile methods and Hidden Markov Models [30,31,32,33,34]. Another kind of a search is based on evaluating the compatibility between the target sequence and each of the structures in the database, achieved by the ``threading'' group of methods [35,36,37,38,39,40]. Threading uses sequence-structure fitness functions, such as residue-level statistical potential functions, to evaluate a sequence-structure match. Threading methods generally do not rely on sequence similarity. Threading sometimes detects structural similarity between proteins without detectable sequence similarity [41].
A good starting point for template searches are the many database search servers on the Internet (Table 2). The most useful ones are those that search directly against the PDB, such as PDB-Blast (http://bioinformatics.burnham-inst.orgpdb_blast). When the target sequence is only remotely related to known structures, it is frequently useful to try several different methods for finding related structures.