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

Integrated Servers for Structure-Based Function Prediction
ProKnow

The first of the two integrated servers described here is ProKnow (Pal and Eisenberg 2005), developed at the University of California, Los Angeles (UCLA) (http://proknow.mbi.ucla.edu). The current version of the server, ProKnow 2.0, employs six main prediction Methods for any uploaded Spatial Structure (Fig. 10.1). In fact, it is also possible to upload just a protein sequence, though in this case one of the six methods cannot be used. The features analyzed by these methods include: overall protein fold, various Structural motifs (omitted if only a sequence is uploaded), sequence similarity, Sequence Motifs, and functional links from the Database of Interacting Proteins (DIP) and the Prolinks database. Each method can provide one or more clues to protein function with varying degrees of confidence. Using Bayes' theorem, weights are assigned to these clues, which are then combined to yield the most probable overall function—expressed in Gene Ontology (GO) terms—along with a confidence score for each. The server outputs a relationship map between the most probable GO Classification predictions (Fig. 10.2), enabling users to interpret the predictions with greater confidence. The final webpage also provides detailed information on the top candidates and their scores. The top candidates for our example, the 2fck structure, are shown in Fig. 10.3. Essentially, the server yields a single best result here: N-acetyltransferase, which is predicted with high confidence and is consistent with the likely function of the protein.