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

Integrated Servers for Structure-Based Function Prediction
Conclusion

Here we have examined ProKnow and ProFunc, two integrated servers that combine sequence- and Structure-matching Methods to predict protein function from a 3D structure uploaded to the server. In most cases, they manage to offer some clues regarding a potential function, although these can sometimes be rather vague (e.g., DNA-binding activity). In other instances, however, all of their underlying methods fail completely and return nothing. The most intriguing scenarios involve structures belonging to uncharacterized families featuring novel fold topologies. Consequently, a researcher might be left knowing only that the structure has an interesting surface cleft lined with highly conserved residues, yet having no idea what might bind within it. Such cases call for The Development of new methods and their integration into existing servers. The most useful would be methods capable of predicting a likely substrate for a given protein based solely on structural analysis. That is, these methods should not rely on matches with existing structures, since, by definition, such matches do not exist for novel folds. Currently, such approaches are highly computationally intensive and typically require at least some prior assumptions regarding the substrate Class (e.g., Hermann et al. 2006). Thus, for some time to come, protein function prediction will continue to rely on clever detective work and logical deduction.