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

Integrated servers for structure-based function prediction
ProKnow
Prediction of fold type

The First stage of ProKnow consists of searching for other Proteins that share the same fold as the protein of interest, or one as closely resembling it as possible. Admittedly, this is somewhat of a shortcut, since ProKnow requires the user to first run the DALI structural alignment program (Holm and Sander 1998) and then upload its output in FSSP format to the server.

Class="center">Structure/structure.files/image068.jpg" width="358"/>

Fig. 10.2. Gene ontology graph generated for the PDB structure 2fck, illustrating the hierarchy of functional terms from general to specific. In cases where ProKnow predicts more than one functional variant, the graph displays a network of these variants, with each connected to others by lines color-coded according to their degree of similarity.

The matches obtained via Dali serve as the initial clues to the protein's function utilized by ProKnow.

Curiously, if only the sequence is uploaded to the server, ProKnow performs the entire workflow autonomously: it identifies a fold compatible with the sequence and uses it as a clue to function. To determine the most probable fold, the ProKnow server relies on the results from the Fold Recognition server developed at UCLA, which also employs a multi-step strategy. First, using the BLAST program, it attempts to find matches between the target sequence and sequences of PDB structures. Next, it tries to obtain a result using the iterative PSI-BLAST program. If both attempts fail, the server utilizes Secondary structure predictions generated by the PSIPRED server maintained at University College London (Bryson et al. 2005). This prediction is fed into the SDP program (Sequence Derived Properties, Fischer and Eisenberg 1997), which attempts to match a suitable packing type. Finally, if this also yields no results, a method known as DASEY (Directional Atomic Solvation EnergY, Mallick et al. 2002) is brought into play.

Fig. 10.3 a) Top functional predictions by the ProKnow server for the PDB structure 2fck. The most confident prediction indicates N-acetyltransferase activity of the protein, b) Summary Table of the clues used to predict each GO term for 2fck. Clicking on any number reveals detailed information regarding that specific clue.

When relying on the results of any fold recognition or threading method, one crucial point must be kept in mind: these Methods operate somewhat like black magic and require cautious interpretation. At times, they can yield a roughly correct answer—typically in the case of small single-domain proteins, which yield topologically near-correct models (Moult 2005); however, overall accuracy varies widely. If the protein sequence is very long, the chances of success drop even further, as the protein almost certainly consists of multiple Structural domains whose boundaries would ideally be defined manually. In such cases, the fold of each domain must be recognized; yet even if these steps are successful, the spatial arrangement of the domains may prove critical to protein function, and domain packing prediction methods have not yet reached maturity (Wollacott et al. 2007; Berrondo et al. 2008).