A Comparison of Various Optimization Algorithms of Protein–ligand Docking Programs by Fitness Accuracy

Liyong Guo,Zhiqiang Yan,Xiliang Zheng,Liang Hu,Yongliang Yang,Jin Wang

DOI: https://doi.org/10.1007/s00894-014-2251-3

IF: 2.172

2014-01-01

Journal of Molecular Modeling

Abstract:In protein–ligand docking, an optimization algorithm is used to find the best binding pose of a ligand against a protein target. This algorithm plays a vital role in determining the docking accuracy. To evaluate the relative performance of different optimization algorithms and provide guidance for real applications, we performed a comparative study on six efficient optimization algorithms, containing two evolutionary algorithm (EA)-based optimizers (LGA, DockDE) and four particle swarm optimization (PSO)-based optimizers (SODock, varCPSO, varCPSO-ls, FIPSDock), which were implemented into the protein–ligand docking program AutoDock. We unified the objective functions by applying the same scoring function, and built a new fitness accuracy as the evaluation criterion that incorporates optimization accuracy, robustness, and efficiency. The varCPSO and varCPSO-ls algorithms show high efficiency with fast convergence speed. However, their accuracy is not optimal, as they cannot reach very low energies. SODock has the highest accuracy and robustness. In addition, SODock shows good performance in efficiency when optimizing drug-like ligands with less than ten rotatable bonds. FIPSDock shows excellent robustness and is close to SODock in accuracy and efficiency. In general, the four PSO-based algorithms show superior performance than the two EA-based algorithms, especially for highly flexible ligands. Our method can be regarded as a reference for the validation of new optimization algorithms in protein–ligand docking.