1/20 Study of Highly Accurate and Fast Protein-Ligand Docking Method Based on Molecular Dynamics Reporter: Yu Lun Kuo Date: November 21, 2006 M. Taufer, M. Crowley, D. J. Price, A. A. Chien‡ and C. L. Brooks III ∗,† Department of Molecular Biology (TPC6), The Scripps Research Institute, North Torrey Pines Road, La Jolla, CA 92037, U.S.A. Published online 24 June 2005 in Wiley InterScience
2/20 Outline Introduction MD-based Docking Method Algorithm Evaluation Metrics MD vs. Other Methods Conclusion & Future work
3/20 Introduction (1/3) Drug development Use of small molecules (ligand) to turn on or off a protein function Protein-ligand docking Computational methods for the prediction of ligand- protein structure information
4/20 Introduction(2/3) Exiting docking method –Current docking algorithms are fast and use simplified scoring function to direct conformational search and select the best structure –Methods based on molecular dynamics (MD) and atomically detailed force field (e.g., CDOCKER) are more accurate but time- and resource-expensive.
5/20 Introduction (3/3) Desktop grids –By scavenging for available and idle cycles –Provide computing power at a significant cost saving Our algorithm –Parallel and each simulation attempt is decomposable into independent sub-jobs
6/20 MD-based Docking Method (1/2) Goals –Assure accuracy Benefit from the molecular mechanics force fields –Guarantee performance Return docking results in a short turnaround time using cost-effective platforms Approach –Docking method based on CHARMM molecular dynamics simulations and with a highly flexible computational granularity
7/20 MD-based Docking Method (2/2) MD Simulation Heating & Cooling phase (300K 700K 300K) Scoring function to rank Lowest energy structure 20 Docking trial
8/20 Algorithm Evaluation (1/2) Characterization of the docking method: –Does the MD length affect the docking accuracy? –Does the number of trials affect the docking accuracy? Comparing algorithm with other well-known docking methods –AutoDock 、 DOCK 、 FlexX 、 ICM 、 GOLD
9/20 Algorithm Evaluation (2/2) Experimental testbed –Platform SGI R10000 –Single 195MHz IP2 processor –128MB memory A cluster of 64 dual-processor nodes at the SDSC (San Diego Supercomputer Center) –Data set: 31 protein-ligand complexes 10 proteins 31 ligands with different levels of complexity
10/20 Metrics Docking Accuracy (DA) –DA = f RMSD< (f RMSD<3 - f RMSD<2 ) – f RMSD<a fraction of predicted ligands docked into a given protein with RMSD lesser or equal to a Ǻ Computational Time –Time to complete a set of docking trials –Report CPU time for sets of 1, 10 and 20 trials
11/20 Four Different MD Simulations
12/20 Docking Accuracy (DA) Ten trials per attempt ensure enough accuracy (T10)
13/20 Average Time with Different Number of MD Steps Increase of number of MD steps Almost linear increase of the simulation time
14/20 MD vs. Other Methods Other Methods –AutoDock 、 DOCK 、 FlexX 、 ICM 、 GOLD Comparison Metrics –Docking Accuracy (DA) –RMSD of predicted ligands –CPU time per attempt Definition of attempt –Consider CASE B and 10 trials per attempt (T10)
15/20 Comparison of Docking Accuracy
16/20 Best RMSD of Predicted Ligands RMSD: Root-mean-square-deviation
17/20 Time Comparison If enough processors are available, the time for completing a protein-ligand docking is competitive with the other methods
18/20 Conclusion & Future work The MD-based docking method –Reach an average accuracy of 71% Still a lot of exciting research has to be addressed both at the application and system levels –Number of ligand orientations per trial based on resources and node reliability
19/20 Conclusion & Future work Future work –Plan to make a more detailed study of MD and Monte Carlo simulations for the docking process in the near future. ICM running multiple Monte Carlo minimizations Our docking protocol to desktop grids –Proportionally decreases the time to solution –Fine-grained parallel algorithm for docking trial
20/20 Thanks for your attention