Herein, we present the results of a machine learning approach we developed to single out correct 3D docking models of protein-protein complexes obtained by popular docking software. To this aim, we generated 3 × 104 docking models for each of the 230 complexes in the protein-protein benchmark, version 5 (BM5), using three different docking programs (HADDOCK, FTDock and ZDOCK), for a cumulative set of ≈ 7 × 106 docking models. Three different machine-learning approaches (Random Forest, Supported Vector Machine and Perceptron) were used to train classifiers with 158 different scoring functions (features). The Random Forest algorithm outperformed the other two algorithms and was selected for further optimization. Using a features selection algorithm, and optimizing the random forest hyperparameters, allowed us to train and validate a random forest classifier, named CoDES (COnservation Driven Expert System). Testing of CoDES on independent datasets, as well as results of its comparative performance with machine-learning methods recently developed in the field for the scoring of docking decoys, confirm its state-of-the-art ability to discriminate correct from incorrect decoys both in terms of global parameters and in terms of decoys ranked at the top positions.
Bibliographical noteKAUST Repository Item: Exported on 2021-12-14
Acknowledgements: The IRaPPA dataset was a courtesy of the methods authors Iain H. Moal and Juan Fernandez-Recio. LC thanks the Supercomputing Laboratory at the King Abdullah University of Science and Technology (KAUST) for technical support and access to the Shaheen facilities. DBB was supported by funding from the AI Initiative at KAUST.
FingerprintDive into the research topics of 'A Random Forest Classifier for Protein-Protein Docking Models'. Together they form a unique fingerprint.