Protein Structure and Function. Application of Bioinformatics Methods - John Rigden 2014

Fold Recognition
Web Tools for Fold Recognition

A wide variety of Structure/29.html">Fold Recognition systems are freely available online for academic use. Some Examples of such systems are listed in Table 2.2. In the recent CASP7 assessment, the I-TASSER, HHpred, Roberta, and Peons Methods demonstrated high performance. Peons, Bioinfo, and Genesilico are meta-servers (or consensus servers) that aggregate modeling results from independent servers and process the generated models using structural clustering or machine learning techniques. Meta-servers generally outperform any individual standalone server. The Roberta server, developed in David Baker's laboratory, is not limited to fold recognition and can generate a broad spectrum of Cell/13.html">Protein Structure predictions, ranging from comparative models to ab initio models. Although the recently developed I-TASSER server was originally designed for fold recognition, the CASP7 meeting demonstrated some encouraging results using I-TASSER for ab initio modeling as well.

Class="center">Table 2.2. Popular web servers for remote Homology/fold recognition. “Consensus” indicates that the server uses results from various independent servers to generate a combined prediction; “Single” indicates that the server relies solely on its own local methods. The “Model Building / Confidence score” Column indicates whether the server generates an output file containing the 3D coordinates of a potential model (“Model”), as well as a score to assess model reliability (such as Z-scores, P-values, E-values, etc.). The “FR / ab initio” column specifies the method: “FR” if the results are based on remote homology/fold recognition methods, or “ab initio” if template-free model building is additionally performed.

The turnaround time for a complete modeling cycle on most of these servers is typically under an hour, with the important caveat that Processing time heavily depends on the current queue load. The ease of use and result interpretation varies widely, and the practical utility of the results for any given user largely depends on their research experience. Furthermore, when tackling prediction tasks, it is always advisable to use multiple servers to rule out false positives.