27
Computational Computational strategies and strategies and methods for building methods for building drug-like libraries drug-like libraries Tim Mitchell, John Holland and John Woods Cambridge Discovery Chemistry & Oxford Molecular

Computational strategies and methods for building drug-like libraries Tim Mitchell, John Holland and John Woods Cambridge Discovery Chemistry & Oxford

Embed Size (px)

Citation preview

Page 1: Computational strategies and methods for building drug-like libraries Tim Mitchell, John Holland and John Woods Cambridge Discovery Chemistry & Oxford

Computational strategies Computational strategies and methods for building and methods for building drug-like librariesdrug-like libraries

Tim Mitchell, John Holland and John Woods

Cambridge Discovery Chemistry & Oxford Molecular

Page 2: Computational strategies and methods for building drug-like libraries Tim Mitchell, John Holland and John Woods Cambridge Discovery Chemistry & Oxford

2

Computational strategies and methods Computational strategies and methods for building drug-like librariesfor building drug-like libraries

What makes a molecule “drug-like” ?

Drug-like screening libraries from commercial sources

Reagent selection

Combinatorial library design

Page 3: Computational strategies and methods for building drug-like libraries Tim Mitchell, John Holland and John Woods Cambridge Discovery Chemistry & Oxford

3

Drug-like propertiesDrug-like properties

Solubility, bio-availability

- Mw, LogP, H-bonds

Toxicity, reactivity

- Topkat

Relatively quick and easy to calculate

- Robust desk-top access can be an issue

Page 4: Computational strategies and methods for building drug-like libraries Tim Mitchell, John Holland and John Woods Cambridge Discovery Chemistry & Oxford

4

Quantitative structure-toxicity Quantitative structure-toxicity relationshipsrelationships

T: Measure of toxicity

- LOAEL, Carcinogenicity, LD50, etc.

A (Pre-exponential factor): Transport quantifiers

- Shape (), Symmetry (S)

G (Free energy term): Electronic properties

- Atomic charges, E-state indices

log (1/[T*log (1/[T*ii]) = log A]) = log Aii - ( - (GGii/2.303 RT) + /2.303 RT) +

logKlogK

Kier, Quant. Struct.-Act. Relat., 5, 1-7 (1986)Kier, Quant. Struct.-Act. Relat., 5, 1-7 (1986)Gombar and Jain, Indian J. Chem., 26A, 554-55 (1987)Gombar and Jain, Indian J. Chem., 26A, 554-55 (1987)Hall et al., J. Chem. Inf. Comput. Sci., 31, 76-82 (1991)Hall et al., J. Chem. Inf. Comput. Sci., 31, 76-82 (1991)

Page 5: Computational strategies and methods for building drug-like libraries Tim Mitchell, John Holland and John Woods Cambridge Discovery Chemistry & Oxford

5

Example Representation of OPSExample Representation of OPS

X2

X10

RRaannggee

ooff

XX22 QQ

Query

Optimum Optimum Prediction Prediction

SpaceSpace(OPS)(OPS)

Range of XRange of X11

W E

I G

H T

W E

I G

H T

H E I G H TH E I G H T

Page 6: Computational strategies and methods for building drug-like libraries Tim Mitchell, John Holland and John Woods Cambridge Discovery Chemistry & Oxford

6

Diamond DiscoveryDiamond DiscoveryTMTM Property Property Calculation & StorageCalculation & Storage

Diamond Calculation ManagerDiamond Calculation Manager

DatabaseDatabasehosthost

Compute Compute serversservers

DesktopDesktopclientsclients

DiamondDiamondPropertiesProperties

DiamondDiamondPharmacophoresPharmacophores

DiamondDiamondToxicityToxicity

TsarTsar DivaDiva ExcelExcel

DiamondDiamondDescriptorsDescriptors

Screening dataScreening dataPredicted dataPredicted dataInventory dataInventory data

John Holland Richard Postance Steve Moon

Page 7: Computational strategies and methods for building drug-like libraries Tim Mitchell, John Holland and John Woods Cambridge Discovery Chemistry & Oxford

7

Core Library Compound SelectionCore Library Compound Selection

Identify ~15,000 compounds from the ~425,000 compounds in our database of commercially available suppliers

Previous experience of Maybridge, BioNet, Menai Organics, AsInEx, ChemStar, Contact Service & Specs indicates their compounds are what they say they are and are >80% pure

Page 8: Computational strategies and methods for building drug-like libraries Tim Mitchell, John Holland and John Woods Cambridge Discovery Chemistry & Oxford

8

Screening Library SelectionScreening Library Selection

Remove unsuitable compounds using calculated properties

- Mol wt. between 200 and 600

- ALogP between -2 and 6

- Estimated LD50 > 100 mg/kg (removes reactive compounds)

- Estimated Ames mutagenicity probability <0.9 (removed hyper-conjugated and activated aromatic)

- Rotatable bonds <= 12

- Likely to be insoluble in 10% DMSO/Water

Cluster on atom & bond fingerprint and select representatives

Visually inspect

Page 9: Computational strategies and methods for building drug-like libraries Tim Mitchell, John Holland and John Woods Cambridge Discovery Chemistry & Oxford

9

Property Based Compound SelectionProperty Based Compound Selection

Page 10: Computational strategies and methods for building drug-like libraries Tim Mitchell, John Holland and John Woods Cambridge Discovery Chemistry & Oxford

10

Core Library Compound SelectionCore Library Compound Selection

All Structures

Preferred suppliers

Mw, LogP, H-BondRot Bond

Ames, LD50

Solubility

- LogP < 3.5

- 3.5 < LogP <4.7& #Ar6 rings <3

425K425K

265K265K

133K133K

89K89K

78K78K

20K20K

19K19K

15K15KStockStock

Page 11: Computational strategies and methods for building drug-like libraries Tim Mitchell, John Holland and John Woods Cambridge Discovery Chemistry & Oxford

11

Screening Library Property ProfilesScreening Library Property Profiles

Mean 33580% 246-427

Mean 2.580% 0.6-4.1

Page 12: Computational strategies and methods for building drug-like libraries Tim Mitchell, John Holland and John Woods Cambridge Discovery Chemistry & Oxford

12

Screening Library Property ProfilesScreening Library Property Profiles

Mean 5.4 Mean 1.1 Mean 3.3

Page 13: Computational strategies and methods for building drug-like libraries Tim Mitchell, John Holland and John Woods Cambridge Discovery Chemistry & Oxford

13

15K Compound Screening Library

- Drug-like

- Non toxic/reactive

- Enhanced solubility

- Diverse

- Visually checked

Samples available for collaborators

- 2mg / well

- 80 compounds / plate

Screening Library from Commercial Screening Library from Commercial SourcesSources

Page 14: Computational strategies and methods for building drug-like libraries Tim Mitchell, John Holland and John Woods Cambridge Discovery Chemistry & Oxford

14

Structure & property-based reagent Structure & property-based reagent selectionselection

Customer request to include -Ph cinnamaldehyde

- Unsuitable for chemistry (reductive amination)

- Suggest alternatives

- Similarity 166 hits, 9 aldehydes

- Substructure + property 47 hits, 47 aldehydes

O

H

MR = 67 AlogP = 3.5# Ar6 = 2

Page 15: Computational strategies and methods for building drug-like libraries Tim Mitchell, John Holland and John Woods Cambridge Discovery Chemistry & Oxford

15

Structure & property-based reagent Structure & property-based reagent selectionselection

Page 16: Computational strategies and methods for building drug-like libraries Tim Mitchell, John Holland and John Woods Cambridge Discovery Chemistry & Oxford

16

Structure & property-based reagent Structure & property-based reagent selectionselection

Page 17: Computational strategies and methods for building drug-like libraries Tim Mitchell, John Holland and John Woods Cambridge Discovery Chemistry & Oxford

17

Library design strategiesLibrary design strategies

Focused library design: Reagent-based selection

- Maximum diversity is not required in focused libraries Systematically optimise substituents

- Synthesise fully enumerated libraries Difficult to cherry-pick and fully enumerate

Reagent selection is compatible with plate layout (8x12 etc.)

- We never know everything about a target Some diversity always necessary

Diverse library design: Product-based selection

- Balance of diversity vs. practical issues

- Product based reagent selection

- 2-D fingerprint / 3-D pharmacophore / 3-D similarity

Drug like properties become increasingly more important as a project progresses from lead discovery to lead optimisation

Page 18: Computational strategies and methods for building drug-like libraries Tim Mitchell, John Holland and John Woods Cambridge Discovery Chemistry & Oxford

18

Library enumeration & profilingLibrary enumeration & profiling

SD file of enumerated library

- Calculate properties (TSAR, Batch TSAR, Diamond Discovery)

Direct calculation from SD file / RS3 Database

Mol wt., Log P, H-bond donors & acceptors

Toxicity

- Analyse profiles (DIVA) Replace any “problem” reagents

- Check for pharmacophores (Chem-X)

- Register as “Work in Progress”

Page 19: Computational strategies and methods for building drug-like libraries Tim Mitchell, John Holland and John Woods Cambridge Discovery Chemistry & Oxford

19

Precursor and property based virtual Precursor and property based virtual library selectionlibrary selection

Register the ID’s of the precursors associated with each product

Select reagent combinations and/or property ranges from large virtual libraries

Page 20: Computational strategies and methods for building drug-like libraries Tim Mitchell, John Holland and John Woods Cambridge Discovery Chemistry & Oxford

20

Library Profiles (DIVA)Library Profiles (DIVA) Rapidly identify precursors which result in undesirable

product properties

Page 21: Computational strategies and methods for building drug-like libraries Tim Mitchell, John Holland and John Woods Cambridge Discovery Chemistry & Oxford

21

Product-based reagent selectionProduct-based reagent selection

Select reagent sub-set and maintain product diversity

Page 22: Computational strategies and methods for building drug-like libraries Tim Mitchell, John Holland and John Woods Cambridge Discovery Chemistry & Oxford

22

Sulfonamide - hydroxamate virtual Sulfonamide - hydroxamate virtual librarylibrary

NH2

O

OS

O

O

Cl

H

H

Br

11 tBu-amino acids

94 sulfonylchlorides

68 benzyl bromides

70,312 virtual products from available reagents

HO

HN

NS

O

R1

R3

O

O

R2

Caldarelli, Habermann & LeyBioorg & Med Chem Lett9 (1999) 2049-2052

Page 23: Computational strategies and methods for building drug-like libraries Tim Mitchell, John Holland and John Woods Cambridge Discovery Chemistry & Oxford

23

Reagent selection & enumerationReagent selection & enumeration

Reject high molecular wt., reactivity

Enumerate 24K products (Afferent)

Calculate product properties (Tsar)

- Mol wt, AlogP

- Estimated Tox. (LD50, Ames)

- Diversity

Profile & select (Diva)

R1 = 11 R2 = 94 R3 = 68R1 = 9 R2 = 40 R3 = 68

NH2

O

O SO

O

ClH

H

Br

Greg Pearl

Page 24: Computational strategies and methods for building drug-like libraries Tim Mitchell, John Holland and John Woods Cambridge Discovery Chemistry & Oxford

24

Virtual Library Profile (Diversity)Virtual Library Profile (Diversity)

Mol Wt. AlogP LogLD50 Cluster R1 R2 R3

Page 25: Computational strategies and methods for building drug-like libraries Tim Mitchell, John Holland and John Woods Cambridge Discovery Chemistry & Oxford

25

Virtual Library Profile (Toxicity)Virtual Library Profile (Toxicity)

Mol Wt. AlogP LogLD50 Cluster R1 R2 R3

Page 26: Computational strategies and methods for building drug-like libraries Tim Mitchell, John Holland and John Woods Cambridge Discovery Chemistry & Oxford

26

Reagent screen & Reagent screen & virtual library profile virtual library profile

Screen reagents- 70,312 (11x94x68) 24,480 (9x40x68)

Reduce Virtual Lib / Maintain Diversity - 24,480 (9x40x68) 8,160 (3x40x68)

Remove likely toxic compounds - 8,160 (3x40x68) 6549 (3x37x59)

Page 27: Computational strategies and methods for building drug-like libraries Tim Mitchell, John Holland and John Woods Cambridge Discovery Chemistry & Oxford

27

Computational strategies and methods Computational strategies and methods for building drug-like librariesfor building drug-like libraries

The ability to calculate, store and search descriptors of hundreds of thousands of compounds is key to both compound selection and library design

Estimated toxicity calculations are useful additions to “standard” molecular descriptors

Calculated properties and analysis tools are readily accessible from a chemists desktop

Property and diversity profiles are very effective, and ensure chemists buy-in to the design process

Oxford Molecular / Cambridge Discover ChemistryBooth 737-740