Structure determination
Structure prediction servers: rise of the machines
Protein structure is increasingly a preoccupation of computers, not of humans. Within the past decade, numerous automated servers have been developed that have taken computational structure prediction to new levels. Such servers, especially those for close and distant homology modelling, are now standard tools for many biomedical research laboratories, but often the question arises as to which server to use for which application.
Since the mid-1990s, the number of automated methods for structural biology has multiplied several-fold, driven in part by the explosive growth of genomic sequence information, but also by the development of increasingly accurate protein structure prediction methods and of course the availability of affordable computing power. Meta-servers, which integrate outputs of various autonomous servers to generate consensus results, are proving to be the most accurate predictors of protein structure and are seen as the new workhorses for most applications, including the prediction of secondary structure, contacts, and docking, as well as close and distant homology modelling.
Daniel Fischer, of the Buffalo Center of Excellence in Bioinformatics in New York, as well as Ben Gurion University in Israel, recently published an informative review in Current Opinion in Structural Biology that evaluates several publicly accessible servers for protein structure prediction. Focusing on resources for close and distant homology modelling, this overview also describes the assessment of such servers through LiveBench and CAFASP (Critical Assessment of Fully Automated Structure Prediction) experiments. One noteworthy observation is that improved methods for distant homology modelling, which ultimately also generate three-dimensional models of proteins based on distantly related template structures, will play an increasingly important role for the functional annotation of genomic sequences, particularly of the many novel open reading frames hitherto labelled merely as 'hypothetical' proteins. Servers, start your engines!
Click here for useful links
