Chemoinformatics in Molecular Docking and Drug Discovery
2 The Docking Problem Given: receptor binding pocket and ligand. Task: quickly find correct binding pose. Two critical modules: 1.Search Algorithm 2.Scoring Function
3 Definitions pKd = measures tightness of binding pKi = measures ability to inhibit Mechanisms of action—for instance: – Competitive inhibition (most typical docking case) – Allosteric inhibition (bind to different pocket) – Allosteric activation
4 Challenges Search algorithm – Speed (5M compounds or more) – Local minima – High-dimensional search space Scoring function – Strict control of false positives – Good correlation with pKd – Multiple terms – No consensus – Non-additive effects (solvation, hydrophobic interactions) Note: pKd does not always correspond with activity ADME concerns
5 Examples of Docking Search Algorithms – Genetic Algorithms – Incremental Construction – Fragment Reconstruction – Gradient Descent – Simulated Annealing and other MC Variants – Tiered Scoring Functions fast screening functions slow accurate functions
6 High Dimensionality: Flexibility Most algorithms handle ligand flexibility but do NOT handle receptor flexibility. Iterative Docking to find alternate conformations of the protein – Dock flexible ligand – Minimize receptor holding ligand rigid – Repeat
7 Scoring Function Energy of Interaction (pKd) Electrostatics Van der Waal’s interactions Hydrogen bonds Solvation effects Loss of entropy Active site waters
8 ADME ADME concerns can be more important than bioactivity. Most of these properties are difficult to predict. Absorption Distribution Metabolism Excretion
9 Docking Programs Dock (UCSF) Autodock (Scripps) Glide (Schrodinger) ICM (Molsoft) FRED (Open Eye) Gold, FlexX, etc.
10 Evaluation of Docking Programs Evaluation of library ranking efficacy in virtual screening. J Comput Chem Jan 15;26(1): Evaluation of docking performance: comparative data on docking algorithms. J Med Chem Jan 29;47(3): Impact of scoring functions on enrichment in docking-based virtual screening: an application study on renin inhibitors. J Chem Inf Comput Sci May-Jun;44(3):
11 Cluster Based Computing Trivially parallelizable – Divide ligand input files – Some programs have specific parallel implementations (PVM or MPI implementations,…) Commercial licenses are expensive
12 Consensus Scoring Combining independent scoring functions and docking algorithms can improve results Most common method: sort using the sum of the ranks of component scores More sophisticated methods exist Consensus scoring criteria for improving enrichment in virtual screening. J Chem Inf Model Jul-Aug;45(4):
13 Adding Chemical Informatics Docking results can be improved by using chemical information about the hits. Chemicals which bind the same protein tend to have similar structure. Iterating back and forth between docking and searching large DB. Use other filters and predictive modules (e.g. Lipinski rules) ALGORITHM: 1.Dock and rank a chemical database 2.Create a bayesian model of the fingerprints of the top hits. 3.Re-rank the database based on their likelihood according to the bayesian model Finding More Needles in the Haystack: A Simple and Efficient Method for Improving High-Throughput Docking Results J. Med. Chem., 47 (11), , 2004.
14 Visualization Viewers must be able to scroll through tens or hundreds of small molecule hits Accessible viewers designed for this problem: – VIDA from OpenEye (free for academics) – ViewDock module of Chimera from UCSF (free, open source)
15 Long-term Goal of Drug Discovery LTDD (Low Throughput Drug Design) instead of HTVS (High Throughput Virtual Screening) Common ground: explore virtual space
Drug Discovery Case Study: Tuberculosis
17 Mycobacterium Tuberculosis Very thick, waxy cell wall Tuberculosis
18 Cell wall lipids: Important for pathogen virulence, survival and latency 6 different ACCase subunits, AccD1-6 Acyl-CoA Tuberculosis 7 th cause of death 1 in 3 people have TB Leading AIDS death cause Multi-drug resistant Mycobacterium tuberculosis Homologs of PccB Focus on AccD4-6 >30 C fatty acid The Cell Wall: Key to Pathogen Survival Sugar 10% of genome
19 Tuberculosis (TB): An old foe
20 The White Death Frederic Chopin John Keats
21 TB: still a real threat, because….. Its ability to stay alive Multi-Drug Resistant (Super TB strain)
22 Cell wall lipids: Important for pathogen virulence, survival and latency 6 different ACCase subunits, AccD1-6 Acyl-CoA Substrate specificity for AccD4-6? Tuberculosis 7 th cause of death worldwide 1 in 3 people have TB Leading cause AIDS death Multi-drug resistant Mycobacterium tuberculosis Homologs of PccB Focus on AccD4-6 >30 C fatty acid The Cell Wall: Key to Pathogen Survival Sugar 10% of genome
23 AccD5 Protein Structures AccD4 (3.3 Å)Solved AccD5 (2.9 Å)AccD6 (2.7 Å)
24 Structure of AccD5
25 Structure-Based Drug Design Crystals & Crystal structure 1. High throughput screening Lead compound2. Virtual Screening 3. Combinatorial chemistry Enzyme assay TB ACCase, AccD5
26 The Computational/Experimental Loop Assay Docking Similarity Search
27 Docking Results Diversity set (1990) from NCI
28 NCI (Lead 1) NCI (Lead 2) 300uM 50uM300uM50-100uM
29 Structure-Based Drug Design Identified AccD5 Inhibitors New TB drug lead K I = 4.7 M, K GI = ~50 M T. Lin, M. Melgar, S. J. Swamidass, J. Purdon, T. Tseng, G. Gago, D. Kurth, P. Baldi, H. Gramajo, and S. Tsai. PNAS, 103, 9, , (2006). US Patent pending.
30 Chemical similarity: Docking: Two Strategies
31 AccD5 Enzyme necessary for mycolic acid biosynthesis in M. tuberculosis.