Fragment Library-Based Models

RNA structure prediction models that utilize a database of solved 3-D structures and structure motifs to piece together new structures are for this work considered fragment library-based (FB) models. FB models stem from the field of protein structure prediction the authors point to [24] for a review of these protein structure models. In the RNA structure review by Laing and Schlick [25], the FB model MC-Sym was favored over all other models and techniques, proving that FB models retain pertinence to RNA structure prediction. The typical simulation scheme for FB models is presented in Fig. 15.2, taken for the RNA molecule PDB ID 4JRC [26]. 

Once FB models collect the RNA fragments to utilize in the RNA structure, the fragments are combined and assembled following the RNA sequence. Following this, FB models employ a Monte Carlo 

---

Another important component of the protein dossier is structural information. Do high-resolution structures (either NMR or X-ray) exist that can be used for library generation and hypothesis testing/generation112,41-441 Are these structures with bound ligands In many medicinal chemists eyes, this is the most important of the criteria in fact, to many it is so important that hit follow-up will not be pursued until suitable structures are in hand. Several companies entire business model is based upon X-ray-based screening of fragment libraries. 

Figure 15.2 Generalized flow diagram of fragment-based models, (a] Sequence information for PDB ID 4JRC. The sequence is divided into (b] coarse-grained sequence fragments, which are then (c] aligned to a preassembled structure library, (d] The best-fit structures are chosen and then [e] combined and minimized to form (f] the final three-dimensional structure. Disclaimer this diagram is generalized and not comprehensive to all fragment-based models.

Constraints provide a very direct means to add information to a simulation— simply requiring all generated structures to satisfy certain additional conditions. This approach has been used extensively to generate three-dimensional structures from NMR spectra [52], which provide data in the form of interatomic distances. In principle, if one had enough distance constraints, the problem would be overdetermined and could be solved mathematically with no further information required. It has been shown, however, that the use of knowledge-based simulations based on homologous structures or fragment libraries from the PDB provides more accurate models than constraint-based methods alone [20,53]. 

---