Evaluating Genetic Algorithms in Protein-Ligand Docking
In silico protein-ligand docking is a basic problem in pharmaceutics and bio-informatics research. One of the very few protein-ligand docking software with available source is the Autodock 3.05 of the Scripps Research Institute. Autodock 3.05 uses a Lamar
- PDF / 592,566 Bytes
- 12 Pages / 430 x 660 pts Page_size
- 108 Downloads / 189 Views
act. In silico protein-ligand docking is a basic problem in pharmaceutics and bio-informatics research. One of the very few protein-ligand docking software with available source is the Autodock 3.05 of the Scripps Research Institute. Autodock 3.05 uses a Lamarckian genetic algorithm for global optimization with a Solis-Wets local search strategy. In this work we evaluate the convergence speed and the deviation properties of the solution produced by Autodock with diverse parameter settings. We conclude that the docking energies found by the genetic algorithm have uncomfortably large deviations. We also suggest a method for considerably decreasing the deviation while the number of evaluations will not be increased.
1
Introduction
In silico protein-ligand docking methods are becoming more and more important in searching for new drug candidate molecules because of their speed, economy and increasing reliability. Acquiring one compound (or ligand) for wet-laboratory testing from compound manufacturers costs around $ 100, consequently, without counting the costs of labor, the additional reagents and the protein production, in vitro verifying of the binding of one million compounds against one protein may cost $ 100 million. In silico simulation of the binding by docking methods costs only a small fraction of that amount, and can be completed in 1-2 weeks on a computer cluster of moderate size. The key ingredient of the in silico docking is the docking algorithm. Each docking algorithm optimizes some scoring function for finding the best location and conformation of the ligand molecule near to the surface of the protein. As an input, one must use the three-dimensional coordinates of the protein (usually taken from the Protein Data Bank) and the ligand molecule. As the output, a docking algorithm returns one or more docked conformation of the ligand and the protein, and the corresponding values of the scoring function. 1.1
The AutoDock Docking Software
One of the most widely known docking software with acquirable source code is the AutoDock 3.05 of the Scripps Research Institute [1]. Note, that the source I. M˘ andoiu, R. Sunderraman, and A. Zelikovsky (Eds.): ISBRA 2008, LNBI 4983, pp. 402–413, 2008. c Springer-Verlag Berlin Heidelberg 2008
Evaluating Genetic Algorithms in Protein-Ligand Docking
403
code of such popular docking software as Dock[2], Gold[3], Fred[4], FlexX[5] and many others are not available at all. In AutoDock 3.05 the scoring function is the estimated docking energy of the ligand to the protein. The best docking is the one with the smallest energy. The derivation of the empirical binding free energy function is described in [1], along with a brief historical review of the topic. It is easy to see that the three-dimensional position of any rigid molecule can be described by 6 real variables (three Euler angles plus the position of one of its atoms). If torsion axes are also allowed, then the scoring function should be optimized in a real space of dimension 6 + . To speed up evaluation of the e
Data Loading...
