Fragment Assembly describes the basic Rosetta protocol for de-novo structure prediction. Fragments provided via a fragment library are randomly selected and inserted into the protein backbone.
In this context a fragment is a particular conformation of protein, RNA or DNA backbone. Usually fragments are the conformation of short consecutive monomeric units (e.g., residues, bases ) but the concept has also been generalized to allow description of broken-chain conformations such as residues in two contacting beta-strands[1]. In the former case, the conformation of a fragment is characterized by a set of backbone torsions (e.g., phi, psi omega angles) and in the latter case, the conformation consists of backbone torsions and rigid-body transformations (Jump).
Before the fragment assembly can commence a fragment library has to be constructed. That is for each set of degrees of freedom (e.g., protein backbone torsions, Jumps) where conformational sampling should take place fragments have to be obtained. For the standard de-novo protein structure prediction protocol, we use the fragment picker to select fragments of 3 and 9 residues length from a precomputed master library of high-resolution protein structures (this library is called vall). The fragment_picker relies on the observation that sequence alone already determines the conformation of short stretches of protein backbone[ref]. In CS-Rosetta the quality (i.e., accuracy and precision) of the selected fragments is greatly enhanced by using chemical shift information in addition to the sequence information[ref]. We usually pick 200 fragments per position.
For the beta-jump sampling, fragments are categorized into 4 types according to direction (parallel, anti-parallel) and pleating (inward, outward) of the strand-strand contact. A file in the rosetta-database has ca. 1000 conformations in each category taken from high-resolution protein structures. No further information but the category is used for their selection.
References