berlin.mpg.de

leipzig.de The identification of protein binding sites and the prediction of protein-ligand complexes play a key role in the pharma-ceutical drug design process and many domains of life sci-ences. Computational approaches for protein-ligand dock-ing (or molecular docking) have received increased attention over the last years as they allow inexpensive and fast pre-diction of protein-ligand complexes. Here we introduce the principle of Bee Nest-Site Selection Optimisation (BNSO), which solves optimisation problems using a novel scheme in-spired by the nest-site selection behaviour found in honey-bees. Moreover, the first BNSO algorithm – called Bee-Nest – is proposed and is applied to the protein-docking prob-lem. The performance of Bee-Nest is tested on 173 docking instances from the PDBbind core set and compared to the performance of three reference algorithms. The results show that Bee-Nest could find ligand poses with very good energy levels. Interestingly, the reference Particle Swarm Optimiza-tion (PSO) produces results that are qualitatively closer to wet-lab experimentally derived complexes but of worse en-ergy than Bee-Nest. Our results clearly highlight the supe-rior performance of Bee-Nest in semi-local (regional) opti-mization for the molecular docking problem, and suggests its usefulness in a hybrid strategy