Protein Docking Algorithm Based On X-ray Free Electron Laser Scattering Map

Protein docking plays an important role in biological functions.Even in the very crowded cellular environment,proteins could recognize each other with a high degree of specificity and form complexes to fulfill their functions.In a human body,the proteins on the virus or on the foreign cells could be detected by immune systems with antigen-antibody interactions.By research on the protein docking,the biochemical process can be studied to help the drug development.Thus,people's interests have been aroused to protein docking problems.Prediction of big protein complexes can be hardly solved by traditional experimental methods,given that:1)degeneracy of NMR signals in large protein challenges the model reconstructions;2)some large proteins fail to crystallize as sample in an X-ray crystallography experiment;3)complexes can be flexible when the proteins combine with each other.Now the best approach to solving the protein docking problem is computer simulation.Since the CAPRI(Critical Assessment of Predicted Interactions)contest was hold,many algorithms have been introduced,such as ZDOCK,Rosetta,etc.Based on the physical interactions,geometric fitness,and the knowledge about the binding sites,these algorithms could make predictions about the complexes and rank the results.It is worth noting that integrating experimental data could promote the accuracy of the protein docking.For example,SAXS-ATTRACT-DOCK,using SAXS data,yielding significant improvements compared with the ATTRACT-DOCK.The introduction of XFEL(X-ray Free Electron Laser)brings new methods to reconstruct protein structures.The experiment could capture a tremendous amount of single particle scattering images in room temperature without crystallizing.Because of unknown orientations and phases,mixed conformations and low signal-noise-ratio,the structural information obtained in the experiment is still limited.In 2014,people made a breakthrough in reconstructing RNAi microsponges structure with 17nm 2D pixel resolution.In 2015,the result of 3D reconstructions of individual biological entities(minivirus)emerged with 100 nm resolution.Yet there is a long way to go to achieve a high resolution as X-ray crystallography or Cryo-electron microscopy.We propose a hybrid approach to determine protein complex structures by combining XFEL single particle scattering data with computational docking methods.For the given proteins,we use ZDOCK program to generate a set of possible structures,and then use the Fourier transform to simulate the scattering patterns.After comparing the patterns with the simulated patterns of native structures,we could re-rank the predicted patterns using different scoring functions.Then we analyzed time consumption and accuracy of orientation finding of each function.The mixture of multi-conformation is also studied.With k-NN joint trained by 3 scoring functions,we could distinguish the conformations of RMSD=0.61A and native structures from the dataset,with only 10000 reference scattering images.The results indicate that our scoring functions are highly selective and could measure the similarity between the proteins precisely.In addition,we suggest an iteration algorithm for optimizing low resolution proteins.In summary,our approach based on molecular modeling and XFEL has a satisfying ranking power on protein docking problems.It should find wide applications in data interpretations for solving the structures of the protein complexes.