Benchmarks and algorithms for protein-protein docking

Protein-protein interactions play a critical role in biological and biochemical systems. Understanding these interactions at an atomic level can aid redesign or modulation of the interaction network. Protein-protein docking attempts to predict the configuration of the binding partners in the complexes from their constituents. Recent rounds of the Critical Assessment of PRedicted Interactions (CAPRI), an international blind test of protein docking, indicate that in many cases high resolution protein-protein complex structures can be predicted.;The success of protein-protein docking has been mostly limited to rigid-body docking, while docking with large conformational changes is one of the major challenges for proteins with flexible loops in interfaces. Surface loops are often found in protein-protein interfaces and are prone to change conformation upon binding. Thus knowledge of loop flexibility and its incorporation in the docking process is useful. After a preliminary investigation of protein surface loop flexibility, I implemented the support vector machine (SVM) approach to distinguish mobile loops from stationary loops and achieved a prediction accuracy of 0.735 and an area under the receiver operating characteristic (ROC) curve of 0.76.;Our lab has demonstrated consistent successful results using the rigid-body docking algorithm ZDOCK and the reranking algorithm ZRANK. Based on my participation in CAPRI, it is critical to evaluate consistency between experimental literature information and ZDOCK results to deal with blind targets. In order to address this task, I introduce the concept of Atom Contact Frequencies (ACF) and applied this in CAPRI round 13-19. ACF enables to check agreement between ZDOCK predictions and experimental information from the literature for blind targets. Furthermore, ACF test results with the newly updated protein-protein docking benchmark 3.0 (precision: 0.67) and published CAPRI targets (precision: 0.9) demonstrate that ACF affords an independent assessment of protein-protein interfaces residue prediction.;To conclude, the development of the ACF has significantly improved the reliability of protein-protein docking methods, although its success is limited to rigid-body cases. The surface loop flexibility analysis provides important insights on the limitation of this rigid-body approximation, and the quantitative prediction of the loop mobility will aid the development of docking methods that include protein flexibility.