Protein structure validation software and its applications in evaluating structures generated by structural genomics and by homology modeling

This work describes a systematic approach to evaluate the quality of protein structures, with a focus on NMR structures and homology models. The Protein Structure Validation Software (PSVS) suite integrates results from several previously published structure quality evaluation tools, evaluating different aspects of structure quality, and generates a standard comprehensive report. PSVS reports global quality measures as Z scores, calibrated on a set of high-resolution X-ray crystal structures. Using these tools to compare the quality of structures generated by structural genomics projects and other projects, we show that the two are of comparable quality. However, there is a structure quality gap between high and medium resolution X-ray structures and most NMR structures; using structure refinement methods incorporating a more accurate description of atomic energies can help narrow this difference.;The HOMA (Homology Modeling Automatically) server generates homology models by satisfaction of spatial restraints. HOMA predicts the structure of the query protein given the sequence alignment between the query and template and the template protein structure, presenting the user with an extensive structure validation report. The performance of the method was evaluated by building homology models for groups of homologous proteins and comparing them to their respective known experimental structures. HOMA provided prediction accuracies similar to other commonly used homology modeling software. Further, we could distinguish between homology models with the correct and incorrect fold, and to a lesser extent, correctly folded homology models of different accuracy, using different structure quality evaluation tools.;An objective and consistent method of accuracy for a homology model would provide guidance on whether the homology model is accurate enough to provide detailed structural information required for docking and mutagenesis studies, or whether only shows the rough fold. Using a combination of sequence alignment and structure quality features, we developed statistical models, using Support Vector Machines and logistic regression, to classify homology models generated by HOMA based on their accuracy. In addition to distinguishing between homology models with the correct and incorrect fold, these models provide good discrimination for homology models with ≤ 1.5 A accuracy from correctly folded homology models of lower accuracy.