AccuRefiner: A machine learning guided refinement method for protein-protein docking

One of the major challenges for protein docking methods is to accurately discriminate native-like structures from false positives. It is known that docking methods are often not accurate and further refinement and re-ranking of the results is required to obtain native-like complexes. We introduce AccuRefiner, a novel machine learning based tool for protein-protein docking refinement. Given a putative docked complex, we produce a small set of refined versions with lower RMSD values than the input docked structure, with respect to the native complex. The method employs a unique ranking tool we developed to predict the RMSD of a docked refinement candidate with respect to the native structure. We show that a properly trained backpropagation neural network can accurately predict the RMSD of a docked complex with less than 1.0Å error margin on average, by approximating the complex relationship between a wide set of scoring function terms and the RMSD of a docked structure. We tested our method on 29 different putative docked complexes produced by three docking programs. The results clearly demonstrate that the high accuracy of the ranking tool enables AccuRefiner to consistently choose the refinement candidates with lower RMSD values compared to the coarsely docked input structures.