Development And Application On Protein-Protein Docking Algorithms

Protein–protein interactions play an important role in many biological progresses,such as gene replication,transcription,translation and regulation,signal transduction,immune recognition,cell transport,catalytic reaction in cells and so on.As structure determines function,determination of the three-dimensional(3D)structures of protein–protein complexes is of great significance for understanding the molecular mechanism behind the protein–protein interaction,accelerating drug discovery and development,treating related diseases and thus improving human health and well-being.However,due to the high cost and technical difficulties of experimental methods,protein–protein docking has become an effective computational method for predicting complex structures.Fast Fourier Transform(FFT)-based docking algorithm has attracted much attention since it was introduced because it can perform a global search efficiently and accurately.Shape complementarity from the"lock and key model"is the most important scoring function in FFT-based docking algorithm,while the commonly used GSC(Grid-based Shape Complementarity)and PSC(Pairwise Shape Complementarity)functions only consider the influences of neighboring atoms around a grid point without the remote atoms.Therefore,we have presented LSC(Long-range Shape Complementarity)scoring function to calculate the long-range effects of protein atoms smoothly by a Gaussian function and developed an FFT-based docking program with LSC.The results have shown that our LSC-based docking program performed better than GSC-or PSC-based docking programs.We also integrated the knowledge-based scoring function ITScore PP into the docking algorithm and developed HDOCK-lite,which significantly improved the docking performance.In order to incorporate the SAXS(Small Angle X-ray Scattering)data that contains the size and shape information of the protein–protein complex,we have developed a procedure for checking the quality of the input SAXS data,and a weighted docking score which combines ITScore PP with SAXS energy score function to sort the predicted binding modes.With the help of SAXS data,the docking success rate on test set has been significantly improved.With the development of traditional structural biology and structural genomics,the number of available protein–protein interface structures in PDB(Protein Data Bank)has increased rapidly.The template-based docking method can use the homologous complex/interface templates to construct the complex structures.The prediction is reliable with the template of high homology,but when there is no template,the free docking algorithm is needed to predict the complex structure.Therefore,we have proposed a hybrid docking strategy,which combines the free docking method HDOCK-lite with the template-based docking method.The hybrid docking improved the prediction performance even only using the weak homologous templates.Finally,we have developed HDOCK server on basis of the above docking methods to help researchers predict the protein–protein complex structures with the help of a variety of experimental information and HDOCK is freely available at http://hdock.phys.hust.edu.cn/.Homo-oligomeric proteins is a major subclass of protein–protein complexes,which are formed by multiple identical subunits.Most homo-oligomeric protein assemblies have certain types of symmetry under the drive of energy and evolution.There are two main problems when using the general docking algorithm to predict the assembly structures of homo-oligomers.Firstly,the conformational sampling is not effective without symmetry constraints.Secondly,it may not be able to construct assembly structures with strict symmetry.Therefore,we have developed a special docking algorithm for predicting the homo-oligomer structures with C_n or D_n symmetry on the basis of HDOCK,called HSYMDOCK.When tested on three test sets consisting of a variety of symmetric proteins,HSYMDOCK has shown better performance than other similar docking algorithms and servers.HSYMDOCK server is available at http://huanglab.phys.hust.edu.cn/hsymdock/.To test the developed docking algorithms and strategies in real applications,we have participated in CAPRI(Critical Assessments of PRediction of Interactions)and achieved excellent performance,especially HDOCK was ranked No.1 among all the server groups in Round 46.We also used HDOCK to predict the complex structure related to chemoresistance in colorectal cancer formed by JAK2 and BECN1 and then analyzed the interaction interface.Finally,inspired by the breakthrough in protein structure prediction achieved by deep learning,we have presented a deep learning model based on Res Net to predict residue–residue contacts across homo-oligomeric interfaces,named as Deep Homo,by integrating structural features and multiple sequence alignment features.Deep Homo achieved better performance than other existing methods on the three test sets.By integrating the predicted contacts into HSYMDOCK,the docking performance has been significantly improved on the test set.Deep Homo is free availbale at http://huanglab.phys.hust.edu.cn/Deep Homo/.