Outlines Drug discovery HTS Screening Computational Screening - - PowerPoint PPT Presentation

Modeling HTS Screening for Drug discovery

Ying-Ta Wu

Genomics Research Center, Academia Sinica, Taiwan. e-mail: ywu@gate.sinica.edu.tw

• Drug discovery
• HTS Screening
• Computational Screening
• Structure-based approach
• Compound-based approach
• Large-Scale Virtual Screening on GRID
• Bioactive Compound Profile

Outlines

New Gene New Target New Lead New Candidate New Drug

~14yrs

Drug Discovery

• Disease Targets
• Chemical Entity ------ > drug

Drug Targets

Human Non-Human

parasites, bacteria, viruses, …

Nature review drug discovery

G.L. Patrick An Introduction to Medicinal Chemistry, Oxford University Press, 1995

Drug Development by chemist intuition …

slay the dragon ! ax sword armor dagger

N N R3 O R1 R2

SAR

(structure Activity relationship)

Drug Discovery by chemist intuition …

Target Assay

Candidate

• ptimization cycle

Library – chemical landscape 10100 Screening – speed vs cost How many compounds can a chemist synthesize per years?

• - few hundreds

N N R3 O R1 R2 NH2 O R1 R2 O O H N H2 R3 X R4

+ + how many? 109 Solution: automation and miniaturize. HTS Alternative? VS Time and Cost

Refer to Walters et al. DDT, 3, 160-178 (1998)

Target selected Assay developed HTS HTS hits confirmed Chemistry begins Target structure obtained Development candidate is taken forward Target selected Assay developed HTS HTS hits confirmed Chemistry begins Target structure obtained Development candidate is taken forward Target selected Assay developed HTS HTS hits confirmed Chemistry begins Target structure obtained Development candidate is taken forward Target selected Assay developed HTS HTS hits confirmed Chemistry begins Target structure obtained Development candidate is taken forward

New Paradigm in Drug Discovery

Screen Strategy

Refer to Walters et al. DDT, 3, 160-178 (1998)

Target selected Assay developed HTS HTS hits confirmed Chemistry begins Target structure obtained Development candidate is taken forward Database clustering Similarity analysis/ Virtual screening Homology modeling QSAR Pharmacophores Structure-based design/ lead optimizing 2-4 years library selecting Target selected Assay developed HTS HTS hits confirmed Chemistry begins Target structure obtained Development candidate is taken forward Database clustering Similarity analysis/ Virtual screening Homology modeling QSAR Pharmacophores Structure-based design/ lead optimizing 2-4 years library selecting Target selected Assay developed HTS HTS hits confirmed Chemistry begins Target structure obtained Development candidate is taken forward Database clustering Similarity analysis/ Virtual screening Homology modeling QSAR Pharmacophores Structure-based design/ lead optimizing 2-4 years library selecting Target selected Assay developed HTS HTS hits confirmed Chemistry begins Target structure obtained Development candidate is taken forward Database clustering Similarity analysis/ Virtual screening Homology modeling QSAR Pharmacophores Structure-based design/ lead optimizing 2-4 years library selecting

New Paradigm in Drug Discovery

Example: Broad (random) Screen

library HTS primary hits reconfirmation assay confirmed hits cluster/MCS/mode hit series

cpd 1 cpd 2 cpd 3

SAR/ADME/IP prioritized hits selected hits

cluster 2 single cluster 2 single cluster 1 cluster 3

extended hits repository substructure similarity library HTS primary hits reconfirmation assay confirmed hits cluster/MCS/mode hit series

cpd 1 cpd 2 cpd 3 cpd 1 cpd 1 cpd 2 cpd 2 cpd 3 cpd 3

SAR/ADME/IP prioritized hits selected hits

cluster 2 cluster 2 single cluster 2 single cluster 2 single cluster 1 single cluster 1 cluster 3 cluster 3

extended hits repository substructure similarity

Drug Discovery by Screening Strategy

Strategy

Random Screen (> 1,000,000) Focused Screen (~10,000) Sequential Screen (5000~10,000)

• whole library screening at initial stage
• no biased, novel
• need uHTS if library is large
• low hit rate, cost
• specific collection screening
• manageable, efficient
• need prior bioactive information
• novelty
• representative subset or explore hit series, which

may be recruited after other two screen procedures

• need clustering, data-mining, etc.
• initial selection

Drug Discovery by Screening

Sequential/Focused Screen

virtual screening medicinal chemistry HTS data analysis active model new library Initial library lead opt

• SD docking
• LB filtering, similar searching
• subset diversity (features)
• clustering

θ

HD HA Z

Information drives drug discovery

• - the more of it, the sooner, the better.

Computational modeling

When to apply What methods and tools virtual library, cluster, screen, or score…

Examples: computational methods and tools

• Target structure-based
• Compound-based
• Score and cluster

Develop inhibitor of SARS coronavirus main protease-3CLpro

Study Case

NCI 25K compounds commercial available structural diversity 2K compounds

A: Pre-screening

modified RO5

create energy maps/element on active-site evaluate 1.5 x 106 energies/molecule carry out 50 x 10 runs/molecule cluster on RMSD=1Å and maxGbinding test compounds with

• Gbinding > 12 kcal/mol

B: Docking Methods

• 1. Prepare the Target Protein
• - add polar hydrogen atoms
• - assign charges to atoms
• - decide range of binding site
• 2. Run AutoGrid
• 3. Prepare the Ligand
• - assign charges to atoms
• - decide flexible bonds (run AutoTors)
• 4. Run AutoDock
• 5. Evaluate Results and Rank Score

Garrett M. Morris David S. Goodsell Ruth Huey William E. Hart Scott Halliday Rik Belew Arthur J. Olson

Morris et al. (1998), J. Computational Chemistry , 19 : 1639-1662.

Docking Engine: AutoDock 3.0.5

S N N HN H NH O HO 26a

• 13 kcal/mol

3x10^ -10 O

S N N HN H NH O HO 26a

• 13 kcal/mol

3x10^ -10 O

R1 R2

Cys-145

Structure-Based Design: Modeling Thiazin derivatives Example:

Glu_166

R3

S N H N COOCH3 40 uM 10 S N H N H N JMF310 10 uM S N H N H N S N JMF311 4 uM S N H N CN JMF312 30 uM S N H N Br JMF313 16 uM S N H N CH3 JMF314 37 uM S N H N CN JMF315 21 uM

S N H N CH3 JMF316 17 uM S N H N OCH3 JMF317 18 uM

S N H N CF3 JMF318 23 uM S N H N CH3 JMF319 > 50 uM

S N H N NO2 JMF320 > 50 uM S N H N JMF321 20 uM S N H N CH3 JMF322 15 uM

S N H N OCH3 JMF323 33 uM S N H N N N N JMF309 > 50 uM

Case Result

Screening by Molecular Docking

Application Characteristics

• Computational screening based on molecular

docking is the most time consuming part in structure-based drug design workflow

• The requirement of CPU power and storage space

increases proportional to the number of compounds and target proteins involved in the screening Number of docking tasks = N x M – N: number of compounds – M: number of target structures

• CPU-bound application, huge amount of output, no

communication between tasks The Challenge

Grid-Enabling Virtual Screening

DCI: against Influenza A

Replication cycle of Flaviviridae

http://en.wikipedia.org/wiki/Aedes Kuhn, R.J.et al. Cell 108, 717−725; 2002

EuAsiaGrid: Dengue virus

Ying-Ta Wu

e-mail: ywu@gate.sinica.edu.tw

GVSS Summary

Grid-enabled Virtual Screening Service (GVSS), incorporating the docking engine of the Autodock 3.0.5, was developed on Grid Application Platform (GAP).

• GVSS uses the DIANE framework to take control of Grid failures and isolate Grid system latency, leading to efficiency,

stability and usability.

• Thanks to DIANE framework and GAP, the GVSS has the ability to efficiently utilize the available Grid resource, which

allows submitted jobs to be split into multiple independent subtasks and run to complete.

• GVSS includes JAVA-based graphical user interface, which allows end-users to specify target and compound library, set up

docking parameters, monitor docking jobs and computing resources, visualize and refine docking results, and finally download the final results. GVSS hides the complexity of deploying large-scale molecular docking on Grid while provides users more flexible control over their works on Grid

• Two applications demonstrated that modeling compound-protein complexes can be speeded up by distributing molecular

docking processes on production Grids. Large-scale compound library can therefore be effectively enriched by executing docking tasks on Grid.

DIANE/AutoDock framework

• 280 DIANE worker agents were submitted as LCG

jobs

• 200 jobs (~71%) were healthy

– ~16 % failures related to middleware errors – ~12 % failures related to application errors

DIANE utilized ~ 95% of the healthy resources

stable throughput

Efficiency and throughput Applications: 1) Anti-Influenza: Data Challenges 2) EUAsiaGrid fight Dengue virus

Supports from Genomics Research Center, Academia Sinica, Taiwan National Science Council, Taiwan and LCG-ARDA, CERN Jakub Moscicki Massimo Lamanna

NS3 protease

cleave polypeptide help virus replication

a high-level grid application framework developed by ASGC

Grid Application Platform (GAP)

Local System Agent (LSA) Virtual Queuing System (VQS)

• 2. prioritized hits with controls

Post-Screening Data Analysis

• 1. Similarity/Activity Clustering
• A lightweight framework for parallel scientific

applications in master-worker model.

• The framework takes care of all synchronization,

communication, and workflow management details on behalf of applications. DIANE, Distributed Analysis Environment

GRID environments

GRID environments

job reassigned a DIANE/Autodock task

• n Grid side
• Due to the loose coupling nature of the Grid:

– Need extra works to manage the efficient job handling and result gathering – Need efforts to handle transient network or site problems – Need application oriented GUI to hide Grid complexities from end users.

• Due to the significant system overhead:

– Grid only benefit to those jobs with long computing time. – not suitable for pilot jobs (required for decision making).

• n Application side
• In the molecular docking based screening, the requirement of CPU power and

storage space increases proportional to the number of compounds (N) and target proteins (M).  Number of docking tasks = N x M

• CPU-bound application, huge amount of output, no communication between tasks
• Task complexity is unpredictable

– difficult to apply trivial domain decomposition method in splitting the tasks

GVSS: Grid-enabling Virtual Screening Service

Molecular docking method is commonly used to predict potential interacting complexes of a small molecule and a target protein. Using molecular docking method for compound screening purpose, however, is restricted by the availability of computing resources. In this work, Grid Application Platform (GAP) and GAP Virtual Screening Service (GVSS) were developed to enable users to get access to the Grid technology and worldwide-scale computing resources seamlessly. Working with production e-infrastructures (such as EGEE and EUAsiaGrid), GVSS presents intensive computing power and effective data management, which provides opportunities for in-silico drug discovery on the neglected and emerging diseases, for instance, Avian Influenza and Dengue Fever. References

Autodock: Morris, G.M., et al., J. Computational Chemistry, 19, 1639-1662 (1998). GAP/GVSS: Lee, H.-C., et al., IEEE Transaction on Nanobioscience, 5, 288-295 (2006) WISDOM: Jacq, N., et al., Parallel Computing, 33, 289-301 (2007) DIANE: Moscicki, J.T., et al., Computer Physics Communications, 180, 2303-2316 (2009) AMGA: Koblitz, B., et al., J. Grid Computing, 6, 61-76 (2008)

GVSS

good load balance

• line segment represents one task=one docking.
• unhealthy workers were removed from the worker list
• failed tasks were rescheduled to healthy workers

Pitfalls Acknowledgements

ASGC, Taiwan His-Kai Wang Mason Hsung* Li-Yung Ho* Hurng-Chun Lee* Wei-Long Ueng Hsing-Yen Chen Eric Yen Simon Lin

Informatics Implications

• Need to be able to store chemical structure and

biological data for millions of datapoints

– Computational representation of 2D structure

• Need to be able to organize thousands of active

compounds into meaningful groups

– Use cluster analysis or machine learning methods to group similar structures together and relate to activity

• Need to learn as much information as possible from the

data (data mining)

– Apply statistical methods to the structures and related information

post-screen data analysis

Extract structural and interacting information To give suggestion :

• Lead structures
• Optimization strategy

Similarity Clustering

Peach and Nicklaus Journal of Cheminformatics 2009, 1:6

Prioritized Filter

Using a Prioritized filter

Rescoring

2qwf (G20) 2qwe (GNA) 1f8b (DAN) 1f8c (4AM) 1f8e (49A) 2qwh (G39)

Control compounds

Applied to validate screening quality and decide the hit rate

prioritized hits with controls

Optimization

1,5-dihydro-2H-pyrrol-2-one

Shie et al. J. AM. CHEM. SOC. 2007, 129, 11892-11893

Evaluation of docked poses Data Analysis

• As a complementary to bio-assay, computational

modeling enable the evaluation of a large compound library.

Summary Computational Screening for Drug Discovery

Accuracy and validity of the computational method is much important than integration at present stage.

• Grid computation is proved to increase the screening speed

enable complete screening of massive compounds.

• GUI make the use of Grid computing resources easier.

Bioactive Compound Profiling Enable the prediction of potential synergic effects or side-effects. Given a search compound…, return with potential interacting targets. T1 T2 T3 T4 T5 C1 C2 C3 C4

Acknowledgements

ASGC, Taiwan Hsin-Yen Chen Eric Yen

• Dr. Simon Lin

His-Kai Wang Mason Hsung* Li-Yung Ho* Hurng-Chun Lee*

Genomics Research Center, Academia Sinica National Science Council, Taiwan