A random forest classifier for protein-protein docking models

Didier Barradas-Bautista; Zhen Cao; Anna Vangone; Romina Oliva; Luigi Cavallo

doi:10.1093/bioadv/vbab042

A random forest classifier for protein-protein docking models

Bioinform Adv. 2021 Dec 10;2(1):vbab042. doi: 10.1093/bioadv/vbab042. eCollection 2022.

Authors

Didier Barradas-Bautista¹, Zhen Cao¹, Anna Vangone², Romina Oliva³, Luigi Cavallo¹

Affiliations

¹ Kaust Catalysis Center, Physical Sciences and Engineering Division, King Abdullah University of Science and Technology (KAUST), 23955-6900 Thuwal, Saudi Arabia.
² Pharma Research and Early Development, Therapeutic Modalities, Roche Innovation Center Munich Large Molecule Research, 82377 Penzberg, Germany.
³ Department of Sciences and Technologies, University Parthenope of Naples, Centro Direzionale Isola C4, I-80143 Naples, Italy.

Abstract

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 $3 \times 10^{4}$ docking models for each of the 230 complexes in the protein-protein benchmark, version 5, using three different docking programs (HADDOCK, FTDock and ZDOCK), for a cumulative set of $\approx 7 \times 10^{6}$ 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 COnservation Driven Expert System (CoDES). 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.

Supplementary information: Supplementary data are available at Bioinformatics Advances online.

Software and data availability statement: The docking models are available at https://doi.org/10.5281/zenodo.4012018. The programs underlying this article will be shared on request to the corresponding authors.