SAIAME: Semi-Parameter Adaptation Information-Assisted Multi-Objective Evolutionary for Protein-Ligand Docking

Abstract

Molecular docking, which simulates the binding pose of a drug molecule to target proteins and predicts the binding affinity, is an important computational tool in structure-based drug discovery. However, the difficulties of high ligand connectivity and dimensionality reduce the search ability of the conformational sampling. To this end, a semi-parameter adaptation information-assisted multi-objective evolution method named SAIAME is proposed for protein-ligand docking optimization. SAIAME employs a staged and dynamic semi-parameter adaptive updating strategy, in which the crossover rate is updated by a weighted arithmetic average algorithm in the exploration phase, as well as the scaling factor is updated by the Lehmer mean in the exploitation phase. It integrates a gradient enhancement based on infinity norms to smooth the decay of the weights of the learning rate during gradient descent to enhance the handling of outliers. It introduces a population size reduction strategy that combines linear and bilateral symmetric sawtooth functions to enhance its execution efficiency. The experimental results demonstrate that SAIAME not only achieves the accuracies of 87.02% for the best poses and 72.98% for the top-score poses within an RMSD of 2 Å, but also has certain advantages in execution efficiency.