Protein structure prediction of CASP5 comparative modeling and fold recognition targets using consensus alignment approach and 3D assessment

For the fifth round of Critical Assessment of Techniques for Protein Structure Prediction (CASP5) all comparative modeling (CM) and fold recognition (FR) target proteins were modeled using a combination of consensus alignment strategy and 3D assessment. A large number and broad variety of prediction targets, with sequence identity between each modeled domain and the related known structure, ranging from 6 to 49%, represented all difficulty levels in comparative modeling and fold recognition. The critical steps in modeling, selection of template(s) and generation of sequence-to-structure alignment, were based on the results of secondary structure prediction and tertiary fold recognition carried out using the Meta Server coupled with the 3D-Jury system. The main idea behind the modeling procedure was to select the most common alignment variants provided by individual servers, as well as to generate several alternatives for questionable regions and to evaluate them in 3D by building corresponding molecular models. Analysis of fold-specific features and sequence conservation patterns for the target family was also widely used at this stage. For both CM and FR targets remote homologs of known structure were clearly recognized by the 3D-Jury system. In the analogous fold recognition subcategory, the correct fold was identified for five out of eight domains. The average alignment accuracy for FR models (48%) was far less than for CM predictions (80%). These finding, coupled with the observation that in the majority of cases the submitted models were not closer to the experimental structure than their best templates, indicate that, especially for difficult targets, there is still ample room for improvement.