Certara, Inc. · NASDAQ:CERT
Certara — Technical Presentation (May 2013)
44 slides from Certara's technical presentation, May 2013
Slide 1: Multi-Criteria Drug Discovery: Recent results in building predictive models
Multi-Criteria Drug Discovery: Recent results in building predictive models, combining predictions, and generating new chemistry ideas Brian B. Masek, Ph.D. Product Manager and Lead Scientist May 2013 1
Slide 2: Today's Presenter
Today’s Presenter • Brian Masek, Ph.D. – Product Manager and Lead Scientist – Development and validation of novel computer-assisted drug discovery methods – Current focus on molecular invention and multi-criteria drug design – Prior to Certara, Director for Scientific R&D at Optive Research and posi
Slide 3: The Problem
The Problem • Compounds exist that have some desirable properties to become a drug but… • The known compounds are not suitable due to: – Intellectual Property – Physical Properties – DMPK Properties – Toxicological Properties – Each with it’s own SAR!! © Copyright 2013 Certara, L.P. All rights reser
Slide 4: Multi-Criteria Drug Discovery: Key Challenges
Multi-Criteria Drug Discovery • Key Challenges: – Generating predictive models for ALL criteria – Visualizing trends across multiple criteria – Prioritizing structures according to multiple criteria – Finding ideas that optimize the balance of properties © Copyright 2013 Certara, L.P. All rights res
Slide 5: Key Challenge: Generating predictive models for ALL criteria
Key Challenge: Generating predictive models for ALL criteria © Copyright 2013 Certara, L.P. All rights reserved.
Slide 6: Safety and Off Target Prediction with 3D QSAR Models
Safety and Off Target Prediction with 3D QSAR Models • Mining public databases helped to solve a problem – Lead compound with potential CYP liability – Neighbours found in PubChem with similar shape-and pharmacophore – 3D-QSAR (Topomer CoMFA) from PubChem derived training set – Model was used to pro
Slide 7: topoHERG – A Classification Model for hERG activity
topoHERG – A Classification Model for hERG activity Nisius, B.; Göller, A.H. Similarity-Based Classifier Using Topomers to Provide a Knowledge Base for hERG Channel Inhibition. J. Chem. Inf. Model.2009, 49, 247-256 © Copyright 2013 Certara, L.P. All rights reserved. 7
Slide 8: Mining Chemogenomic Spaces – the QSARome
Mining Chemogenomic Spaces – the “QSARome” Capturing SAR-Trends and QSAR’s is generally feasible – Chemogenomic databases are usable sources of SAR information – 3D-QSAR models can be obtained through automation • Topomer CoMFA • QSEA – The 3D-QSAR models and the SAR-trends derived from them are hel
Slide 9: Automation of 3D-QSAR for Mining Chemogenomic Spaces
Automation of 3D-QSAR for Mining Chemogenomic Spaces • The challenge: – Traditional 3D-QSAR techniques require 3D alignment – Alignment of training set required to build QSAR model – Alignment of prediction set to make predictions – 3D alignment is subjective • Topomer CoMFA – Fragment based 3D-QSAR
Slide 10: Mining Chemogenomic Spaces – the QSARome
Mining Chemogenomic Spaces – the QSARome Workflow for mining databases for 3D-QSAR’s © Copyright 2013 Certara, L.P. All rights reserved. 10
Slide 11: Mining Chemogenomic Spaces – the QSARome
Mining Chemogenomic Spaces – the QSARome ~50 Million SAR data points ~25,000 Raw 5718 Filtered 1795 3D-QSAR ~5,000 biological SAR Tables SAR Tables Models targets Data Quality • PubChem & ChemBank • HTS-data • chEMBL • Dose/Response Data © Copyright 2013 Certara, L.P. All rights reserved. 11
Slide 12: Mining Chemogenomic Spaces – the QSARome
Mining Chemogenomic Spaces – the QSARome • hERG Channel Model • QSEA: • Similarity Distance Matrix – One major subset • Activity Range – Good mixture of actives and inactives – Range: 0.15-7.0 uM • SAR-Trend detected © Copyright 2013 Certara, L.P. All rights reserved. 12
Slide 13: Key Challenge: Generating predictive models for ALL criteria
Key Challenge: Generating predictive models for ALL criteria • 3D-QSAR can make useful predictions for off-targets and can help solve discovery problems • Capturing SAR-Trends and QSAR’s is generally feasible – Chemogenomic databases are usable sources of SAR information – 3D-QSAR models can be obta
Slide 14: Python Toolkit for 3D-QSAR
Python Toolkit for 3D-QSAR • Create 3D-QSAR models – Traditional CoMFA – Topomer CoMFA • Make predictions with 3D-QSAR models • Python interface to: -------------------------------------------------------------------------------------------------- #!/bin/env python – 3D field generator from tripos.q
Slide 15: Key Challenge: Ranking structures according to multiple criteria
Key Challenge: Ranking structures according to multiple criteria © Copyright 2013 Certara, L.P. All rights reserved.
Slide 16: Scoring Philosophy
Scoring Philosophy • For some properties, bigger is better – E.g Potency (as measured by docking score or ligand similarity) • For some properties, smaller is better – E.g selectivity • For some properties, a range of values is acceptable – E.g. logP, MW, TPSA, Lipinski properties, etc… • Filtering
Slide 17: Designing using Opposing Forces
Designing using Opposing Forces Like Unlike Recognition Features Based on Shape Structural Similarity • Prevents “rediscovery” of the lead compounds used for similarity scoring © Copyright 2013 Certara, L.P. All rights reserved. 17
Slide 18: Structural Similarity for Molecular Invention
Structural Similarity for Molecular Invention Lead Exp Lead Opt • UNITY fingerprint Tanimoto similarities are turned into UNITY scores • Final Score = Raw Score * Unity Score – Structures are penalized if they are too close or too far from the template(s) 2D Fingerprint Score when desired UNITY Tani
Slide 19: Combining Multiple Parameters
Combining Multiple Parameters Pareto Rank • Pareto Rank is a count of the number of structures that “dominate” in all parameters • Many structures can have a Pareto rank of “0” • Weak Dominance important if structures can achieve perfect score Borda Voting • Used to order structures with the same Pa
Slide 20: Normalizing and Combining Multiple Scores
Normalizing and Combining Multiple Scores Weighted Sums: 𝑁 𝑆𝑐𝑜𝑟𝑒 = 𝑤𝑖 ∗ 𝑠𝑖 𝑖=1 Weighted Products: 𝑁 𝑆𝑐𝑜𝑟𝑒 = 𝑤𝑖 ∗ 𝑠𝑖 𝑖=1 © Copyright 2013 Certara, L.P. All rights reserved. 20
Slide 21: © Copyright 2013 Certara, L.P. All rights reserved.
© Copyright 2013 Certara, L.P. All rights reserved.
Slide 22: Key Challenge: Finding ideas that optimize the balance of properties
Key Challenge: Finding ideas that optimize the balance of properties © Copyright 2013 Certara, L.P. All rights reserved.
Slide 23: The Problem
The Problem • Compounds exist that have some desirable properties to become a drug but… • The known compounds are not suitable due to: – Intellectual Property – Physical Properties – DMPK Properties – Toxicological Properties – Each with it’s own SAR!! © Copyright 2013 Certara, L.P. All rights reser
Slide 24: The Design Challenge
The Design Challenge What compound should I make next? • How to generate ideas that will satisfy ALL the criteria for a successful drug candidate? © Copyright 2013 Certara, L.P. All rights reserved. 24
Slide 25: De Novo Design for Multi-Criteria Drug Design
De Novo Design for Multi-Criteria Drug Design For both identification and optimization of lead candidates Identify novel structures, scaffolds, or side-chains that meet specific design objectives. • Lead hopping – invent whole new ideas • Scaffold hopping • Invent new R-Groups around a fixed scaffol
Slide 26: Overview of Molecular Invention Process
Overview of Molecular Invention Process May Preserve Scaffold • R-group Optimization May Preserve R-Groups Seed • Scaffold Hopping Structures Invent Population Population of Parents Of Children (With Scores) Repeat for Number of Generations Result Decouple “Invention from Structures “Scoring” Select
Slide 27: An Example of how an Invention Progresses
An Example of how an Invention Progresses 1 CP99994 2 N ClogP 3.4 HBA 3 N HBD 2 O MW 296 PSA 38 RotBond 5 NK1 Antagonist CP99994 © Copyright 2013 Certara, L.P. All rights reserved. 27
Slide 28: An Example of how an Invention Progresses
An Example of how an Invention Progresses Hypothetical CP99994 Target N ClogP 3.4 4.0-4.5 HBA 3 2 N HBD 2 2 O MW 296 330-370 PSA 38 25-30 RotBond 5 2-3 NK1 Antagonist CP99994 © Copyright 2013 Certara, L.P. All rights reserved. 28
Slide 29: Initial Compound CP99994
Initial Compound CP99994 N N O ClogP RotBond HBA PSA HBD MW Init Target ClogP 3.4 4.0-4.5 HBA 3 2 HBD 2 2 MW 296 330-370 PSA 38 25-30 RotBond 5 2-3 © Copyright 2013 Certara, © Certara, L.P.L.P. 2012All Allrights Rightsreserved. Reserved 29
Slide 30: Initial Compound and Generation 1
Initial Compound Generation 1 CP99994 G001_C001 N N O ClogP ClogP RotBond HBA RotBond HBA PSA HBD PSA HBD MW MW Init Target G1 ClogP 3.4 4.0-4.5 3.9 HBA 3 2 2 HBD 2 2 2 MW 296 330-370 280 PSA 38 25-30 29 RotBond 5 2-3 4 © Copyright 2013 Certara, © Certara, L.P.L.P. 2012All Allrights Rightsreserved.
Slide 31: Generation 3 compound
Initial Compound Generation 1 Generation 3 CP99994 G001_C001 G003_C005 N N O ClogP ClogP ClogP RotBond HBA RotBond HBA RotBond HBA PSA HBD PSA HBD PSA HBD MW MW MW Init Target G1 G3 ClogP 3.4 4.0-4.5 3.9 5.0 HBA 3 2 2 2 HBD 2 2 2 2 MW 296 330-370 280 336 PSA 38 25-30 29 29 RotBond 5 2-3 4 4 © Copyri
Slide 32: Generation 4 compound
Initial Compound Generation 1 Generation 3 CP99994 G001_C001 G003_C005 N N O ClogP ClogP ClogP RotBond HBA RotBond HBA RotBond HBA PSA HBD PSA HBD PSA HBD MW MW MW Init Target G1 G3 G4 Generation 4 ClogP 3.4 4.0-4.5 3.9 5.0 4.5 G004_C006 HBA 3 2 2 2 2 HBD 2 2 2 2 2 MW 296 330-370 280 336 322 ClogP P
Slide 33: Generation 9 compound
Initial Compound Generation 1 Generation 3 CP99994 G001_C001 G003_C005 N N O ClogP ClogP ClogP RotBond HBA RotBond HBA RotBond HBA PSA HBD PSA HBD PSA HBD MW MW MW Init Target G1 G3 G4 G9 Generation 4 Generation 9 ClogP 3.4 4.0-4.5 3.9 5.0 4.5 4.4 G004_C006 G009_C002 HBA 3 2 2 2 2 2 HBD 2 2 2 2 2 2
Slide 34: Final invented compound
Initial Compound Generation 1 Generation 3 CP99994 G001_C001 G003_C005 N N O ClogP ClogP ClogP RotBond HBA RotBond HBA RotBond HBA PSA HBD PSA HBD PSA HBD MW MW MW Init Target G1 G3 G4 G9 Generation 4 Generation 9 ClogP 3.4 4.0-4.5 3.9 5.0 4.5 4.4 G004_C006 G009_C002 HBA 3 2 2 2 2 2 HBD 2 2 2 2 2 2
Slide 35: The Design Puzzle
The Design Puzzle QSAR ADME Structure based Docking Drug Ligand like based Lead Shape based Optimization screening or Lead Filters hopping Pharmacophore Synthesizable © Copyright 2013 Certara, L.P. All rights reserved. 35
Slide 36: Including ADME Scoring - Baseline
Including ADME Scoring - Baseline The p38 ligands are mainly predicted to be permeable by the VolSurf+ Caco2 model © Copyright 2013 Certara, L.P. All rights reserved. 36
Slide 37: ADME Scoring – Inventions without ADME
ADME Scoring – Inventions without ADME Muse invents structures ranging from poor to high permeability © Copyright 2013 Certara, L.P. All rights reserved. 37
Slide 38: ADME Scoring Function – Inventions with ADME
ADME Scoring Function – Inventions with ADME Adding the VolSurf+ Caco2 model to the Muse scoring function - the population is more tightly focused in the permeable region © Copyright 2013 Certara, L.P. All rights reserved. 38
Slide 39: A Success Story - Neurocrine
A Success Story - Neurocrine • Invention of Selective Norephinephrine Re-uptake Inhibitors O O H O N O O O N Training Set O O H O O H N O N H H ES 2131020 WO 04018441 WO 04018441 CF3 N H Muse “Engine” O Atomoxetine Reboxetine O NH NH NH H N O MDL 27777A ES 2131020 DE 2627227 HN US 6759435 HN Cl Mult
Slide 40: More Success - AstraZeneca
More Success - AstraZeneca • Described several examples of successful “retrospective leadhopping” using molecular invention based on multiple criteria (See Webinar at www.tripos.com) • Described design of novel non-basic side chain to solve DMPK/Tox problem MeO H Muse “Engine” H N N O O N O O N N N
Slide 41: Publications
Publications Toluidinesulfonamide Hypoxia-Induced Factor 1 Inhibitors: Alleviating Drug–Drug Interactions through Use of PubChem Data and Comparative Molecular Field Analysis Guided Synthesis. B. Wendt, et.al. J. Med. Chem., 2011, 54 (11), 3982–3986 Similarity-Based Classifier Using Topomers to Prov
Slide 42: Multi-Criteria Drug Discovery Report Card
Multi-Criteria Drug Discovery Report Card • Key Challenges: – Generating predictive models for ALL criteria – Visualize trends across mulitple criteria – Ranking structures according to multiple criteria – Finding ideas that optimize the balance of properties – Automated 3D QSAR Methods can be used
Slide 43: Acknowledgements
Acknowledgements • Certara • AstraZeneca – Richard Cramer – Jim Damewood – Bernd Wendt – Charles Lerman – Fred Soltanshahi – Fabian Boes • Molecular Discovery – Ulrike Uhrig – Simon Cross • Elara – Joe Lewis – Jochen Ammenn – Christoph Schultes – Bernd Janssen © Copyright 2013 Certara, L.P. All righ
Slide 44: Contacts & Follow-Up: Thank you!
Contacts & Follow-Up: Thank you! For more information regarding today’s presentation, please contact: Brian Masek, PhD., Product Manager and Lead Scientist [email protected] For more information on Certara’s products & services, please contact our sales group: Americas India Telephone: +1-919-