Improving the efficiency of PSOVina for protein-ligand docking by two-stage local search

Protein-ligand docking programs are valuable tools in the modern drug discovery process for predicting the complex structure of a small molecule ligand and the target protein. Often, the configurational search algorithm in the docking tool consists of global search and local search. The former is to explore widely for promising regions in the search space and the latter is to optimize a candidate solution to a local optimum. However, accurate local search methods such as gradient-based Newton methods are very costly. In this investigation, we present a new approach to enhance the time efficiency of a docking program by introducing a two-stage local search method. Given a candidate solution, a rough local search is performed in the first stage to determine the potentiality of the solution. Only if the solution is promising, the second stage with a full local search will be performed. Our method has been realized in the PSOVina docking program and tested on two data sets. The experimental results show that two-stage local search achieves almost 2x speedup to conventional one-stage method, it also enhances the prediction performance of the docking method in terms of increased success rate and RMSD.