Building innovative

drug discovery alliances

Algorithms, Evolution and Network-Based Approaches in Molecular Discovery

ChemAxon UGM, Budapest, May2015

PAGE

Drug Discovery

1

The State of the Art

PAGE

Antipsychotic drugs and their Poly-Pharmacology

2

The chart shows the known affinity (Ki) values of antipsychotic drugs for a panel of receptors.

Haloperidol

Loxapine Clozapine OlanzapineZiprasidone

Thioridazine ThiothixeneZotepine

How can we go about discovering a novel antipsychotic?.

PAGE

What do I make next?

3

Medicinal Chemistry Optimisation

PAGE

ChEMBL

4

ChEMBLSpace

. https://www.ebi.ac.uk/chembldb/

ChEMBLSpace – a graphical explorer of the

chemogenomic space covered by ChEMBL

Bioinformatics (2013) 29 (4): 523-524

10.6k Targets

1.4m Compounds

12.8m Activities

PAGE

ChEMBLSpace search: D2 & a1BA

5 Available for download search: ChEMBLSpace@sourceforge.net

Dopamine D2

a1B-Adrenergic

PAGE

ChEMBLSpace: D2, a1BA, H1

6 Available for download search: ChEMBLSpace@sourceforge.net

Histamine H1

Dopamine D2

a1B-Adrenergic

PAGE

Similarity Ensemble Approach (SEA)

7

Keiser et al.

Nature 462, 175-181(12 November 2009)

PAGE

Predictive Pharmacology

8 Drakakis et al., Med Chem Commun 5, 386-396 (2014)

• Combination of: • Target Prediction • BioSAR - Laplacian-modified Naïve Bayes algorithm • Information gain prefiltering• Decision Trees (C4.5) for Classification

• Yields 70% accurate,

• Interpretable model for sleep outcome.

BIOPRINT

IF

ACTIVITY_ON D(2) Dopamine receptor

AND

ACTIVITY_ON Histamine H1 receptor

AND

ACTIVITY_ON 5-hydroxytryptamine receptor 2A

THEN

“Good Sleep”

PAGE

Accessing Chemical Space

9 J. Chem. Inf. Model., 2012, 52, 2864-2875

Heavy Atom Count (hac)

166 Billion

2.5 Million

89 Thousand

367 Compounds

A virtual enumeration of chemical space up to 17 heavy atoms generated 166,443,860,262 molecules *

Plo

tte

d o

n a

lo

g s

ca

le

A Pharma screening collection up to 17 heavy atoms is typically 100–500K molecules

== 0.000003% of accessible space

PAGE

Strategies in Automated Molecule Design

10 De Novo Design 2013 | ISBN 978-3-527-67700-9

Core X BA

C

Core X BA

CGrow

Replace Scaffold

Constraints

2D or 3D

Generation

Protein Cavity

Volume Based

Feature Based

SyntheticFeasibility

Reaction Based

Reagent Enumeration

Transformation Based

Complexity Score

User Assessment

Core X B

C

Core Y

Merge Molecules

A

PAGE

Motivation

11

Using reaction databases

public

reaction

databases

reactions

covering general

organic chemistryreaction

database

commercial

reaction

databasesU

Adding say ~250,000

reactions per year, strong

medicinal chemistry bias

wealth of reaction data

– extract the knowledge hidden in these data

– use this knowledge to assist the medicinal chemist

– suggest new, synthetically feasible molecules with desired bio profile

500 chemists

~10 reactions per week

lab notebooks (eLN)

proprietary

reaction

database

PAGE

Reaction Vectors

12 Broughton, H. B., Hunt, P. & Mackey, M. (2003) Methods for Classifying and Searching Chemical Reactions.

United States Patent Application 367550.

I 1 2 3 4

Bond C-C C=O C-OH C-OR

# 0 0 -2 2

reactant vector, R = (R1 + R2) product vector, P

reaction vector, D = P - R

OH

O

OHO

O

+

I 1 2 3 4

Bond C-C C=O C-OH C-OR

# 4 1 2 0

I 1 2 3 4

Bond C-C C=O C-OH C-OR

# 4 1 0 2

R1 R2 P

PAGE

Reaction Vectors in Structure Generation

13 J. Chem. Inf. Model., 2009, 49 (5), pp 1163–1184. Knowledge-Based Approach to de Novo Design Using Reaction Vectors

• The reaction vector, D, equals the difference between the product

vector, P, and the reactant vector, R

D = P – R

I 1 2 3 4

Bond C-C C=O C-OH C-OR

# 4 1 0 2

O O

O

O

O

O

better descriptor

is required

Given a reaction vector, D, and a reactant vector, R, the product vector, P,

can be obtained

P = D + R

Given a product vector, P, can we reconstruct the product molecule(s)?

PAGE

Modified Atom Pairs

14

Atom Pairs 2 (AP2) : X1(n, p, r)-2(BO)-X2(n, p, r)

X: element type

n: number of bonds to heavy atoms

p: number of π bonds

r: number of ring memberships

BO: bond order

Atom Pairs 3 (AP3): X1(n, p, r)-3-X2(n, p, r)

• Extending the bond distance in atom pairs encodes more of

the environment of the reaction centre

PAGE

Beckmann Rearrangement

15

Cl

N

OH Cl

NH

O

Reaction vector

O(1,1,0)-3-C(1,0,0)f+N(2,1,0)-3-C(1,0,0)f-

N(2,0,0)-3-O(1,1,0)e+C(3,1,0)-3-O(1,0,0)e-

N(2,0,0)-3-C(1,0,0)d+C(3,1,0)-3-C(2,2,1)d-

C(2,2,1)-3-N(2,0,0)c+C(3,1,0)-3-C(2,2,1)c-

C(2,2,1)-3-N(2,0,0)b+C(3,2,1)-3-C(1,0,0)b-

C(2,2,1)-3-N(2,0,0)a+C(3,2,1)-3-N(2,1,0)a-

C(3,1,0)-2(2)-O(1,1,0)3+N(2,1,0)-2(1)-O(1,0,0)3-

C(3,1,0)-2(1)-N(2,0,0)2+C(3,1,0)-2(2)-N(2,1,0)2-

C(3,2,1)-2(1)-N(2,0,0)1+C(3,2,1)-2(1)-C(3,1,0)1-

Positive APsNegative APs

O(1,1,0)-3-C(1,0,0)f+N(2,1,0)-3-C(1,0,0)f-

N(2,0,0)-3-O(1,1,0)e+C(3,1,0)-3-O(1,0,0)e-

N(2,0,0)-3-C(1,0,0)d+C(3,1,0)-3-C(2,2,1)d-

C(2,2,1)-3-N(2,0,0)c+C(3,1,0)-3-C(2,2,1)c-

C(2,2,1)-3-N(2,0,0)b+C(3,2,1)-3-C(1,0,0)b-

C(2,2,1)-3-N(2,0,0)a+C(3,2,1)-3-N(2,1,0)a-

C(3,1,0)-2(2)-O(1,1,0)3+N(2,1,0)-2(1)-O(1,0,0)3-

C(3,1,0)-2(1)-N(2,0,0)2+C(3,1,0)-2(2)-N(2,1,0)2-

C(3,2,1)-2(1)-N(2,0,0)1+C(3,2,1)-2(1)-C(3,1,0)1-

Positive APsNegative APs

Atom Pairs 2 (AP2) : X1(n, p, r)-2(BO)-X2(n, p, r)

Atom Pairs 3 (AP3): X1(n, p, r)-3-X2(n, p, r)

X element type

n number of bonds to heavy atoms

p number of π bonds

r number of ring memberships

BO bond order

PAGE

Applying a RV to a reactant to generate a Product

16

CH3

C5

CH

1

CH2

C6

CH

4

Cl7

C8

CH3

9

O(1,1,0)-3-C(1,0,0)f+N(2,1,0)-3-C(1,0,0)f-

N(2,0,0)-3-O(1,1,0)e+C(3,1,0)-3-O(1,0,0)e-

N(2,0,0)-3-C(1,0,0)d+C(3,1,0)-3-C(2,2,1)d-

C(2,2,1)-3-N(2,0,0)c+C(3,1,0)-3-C(2,2,1)c-

C(2,2,1)-3-N(2,0,0)b+C(3,2,1)-3-C(1,0,0)b-

C(2,2,1)-3-N(2,0,0)a+C(3,2,1)-3-N(2,1,0)a-

C(3,1,0)-2(2)-O(1,1,0)3+N(2,1,0)-2(1)-O(1,0,0)3-

C(3,1,0)-2(1)-N(2,0,0)2+C(3,1,0)-2(2)-N(2,1,0)2-

C(3,2,1)-2(1)-N(2,0,0)1+C(3,2,1)-2(1)-C(3,1,0)1-

Positive APsNegative APs

O(1,1,0)-3-C(1,0,0)f+N(2,1,0)-3-C(1,0,0)f-

N(2,0,0)-3-O(1,1,0)e+C(3,1,0)-3-O(1,0,0)e-

N(2,0,0)-3-C(1,0,0)d+C(3,1,0)-3-C(2,2,1)d-

C(2,2,1)-3-N(2,0,0)c+C(3,1,0)-3-C(2,2,1)c-

C(2,2,1)-3-N(2,0,0)b+C(3,2,1)-3-C(1,0,0)b-

C(2,2,1)-3-N(2,0,0)a+C(3,2,1)-3-N(2,1,0)a-

C(3,1,0)-2(2)-O(1,1,0)3+N(2,1,0)-2(1)-O(1,0,0)3-

C(3,1,0)-2(1)-N(2,0,0)2+C(3,1,0)-2(2)-N(2,1,0)2-

C(3,2,1)-2(1)-N(2,0,0)1+C(3,2,1)-2(1)-C(3,1,0)1-

Positive APsNegative APs

Cl

N

OH Cl

NH

O

1. Removing the negative atom pairs from the reactant

1- 2-

3-

PAGE

Applying a RV to a reactant to generate a Product

17

No AP2s left in the reaction

vector that match atom 11

CH3

5

CH

1

CH2

C6

CH

4

Cl7

C8

CH3

9

CH3

5

CH

1

CH2

6CH

4

Cl7

C8

CH3

9

NH10

CH3

C5

CH

1

CH2

C6

CH

4

Cl7

C8

CH3

9

O10

CH3

C5

CH

1

CH2

C6

CH

4

Cl7

C8

CH3

9

NH

10

C11

CH3

5

CH

1

CH2

6

CH

4

Cl7

NH

10C8

CH3

9

CH3

C5

CH

1

CH2

C6

CH

4

Cl7

C8

CH3

9

O10

NH11

CH3

5

CH

1

CH2

6

CH

4

Cl7

NH

10C 8

CH3

9

O11

CH3

C5

CH

1

CH2

C 6

CH

4

Cl7

C8

CH3

9

O10

NH

11

Duplicate solutionFinal Solution

2+ (d+) 3+ (f+)

1+ (a+)1+ (a+,b+,c+) 2+ (d+,e+)

3+ (e+,f+) 1+ (a+,b+,c+)

A

B C

D E F

G H

No AP2s left in the reaction

vector that match atom 11

CH3

5

CH

1

CH2

C6

CH

4

Cl7

C8

CH3

9

CH3

5

CH

1

CH2

6CH

4

Cl7

C8

CH3

9

NH10

CH3

C5

CH

1

CH2

C6

CH

4

Cl7

C8

CH3

9

O10

CH3

C5

CH

1

CH2

C6

CH

4

Cl7

C8

CH3

9

NH

10

C11

CH3

5

CH

1

CH2

6

CH

4

Cl7

NH

10C8

CH3

9

CH3

C5

CH

1

CH2

C6

CH

4

Cl7

C8

CH3

9

O10

NH11

CH3

5

CH

1

CH2

6

CH

4

Cl7

NH

10C 8

CH3

9

O11

CH3

C5

CH

1

CH2

C 6

CH

4

Cl7

C8

CH3

9

O10

NH

11

Duplicate solutionFinal Solution

2+ (d+) 3+ (f+)

1+ (a+)1+ (a+,b+,c+) 2+ (d+,e+)

3+ (e+,f+) 1+ (a+,b+,c+)

A

B C

D E F

G H

CH3

5

CH

1

CH2

C6

CH

4

Cl7

C8

CH3

9

CH3

5

CH

1

CH2

6CH

4

Cl7

C8

CH3

9

NH10

CH3

C5

CH

1

CH2

C6

CH

4

Cl7

C8

CH3

9

O10

CH3

C5

CH

1

CH2

C6

CH

4

Cl7

C8

CH3

9

NH

10

C11

CH3

5

CH

1

CH2

6

CH

4

Cl7

NH

10C8

CH3

9

CH3

C5

CH

1

CH2

C6

CH

4

Cl7

C8

CH3

9

O10

NH11

CH3

5

CH

1

CH2

6

CH

4

Cl7

NH

10C 8

CH3

9

O11

CH3

C5

CH

1

CH2

C 6

CH

4

Cl7

C8

CH3

9

O10

NH

11

Duplicate solutionFinal Solution

2+ (d+) 3+ (f+)

1+ (a+)1+ (a+,b+,c+) 2+ (d+,e+)

3+ (e+,f+) 1+ (a+,b+,c+)

A

B C

D E F

G H

O(1,1,0)-3-C(1,0,0)f+N(2,1,0)-3-C(1,0,0)f-

N(2,0,0)-3-O(1,1,0)e+C(3,1,0)-3-O(1,0,0)e-

N(2,0,0)-3-C(1,0,0)d+C(3,1,0)-3-C(2,2,1)d-

C(2,2,1)-3-N(2,0,0)c+C(3,1,0)-3-C(2,2,1)c-

C(2,2,1)-3-N(2,0,0)b+C(3,2,1)-3-C(1,0,0)b-

C(2,2,1)-3-N(2,0,0)a+C(3,2,1)-3-N(2,1,0)a-

C(3,1,0)-2(2)-O(1,1,0)3+N(2,1,0)-2(1)-O(1,0,0)3-

C(3,1,0)-2(1)-N(2,0,0)2+C(3,1,0)-2(2)-N(2,1,0)2-

C(3,2,1)-2(1)-N(2,0,0)1+C(3,2,1)-2(1)-C(3,1,0)1-

Positive APsNegative APs

O(1,1,0)-3-C(1,0,0)f+N(2,1,0)-3-C(1,0,0)f-

N(2,0,0)-3-O(1,1,0)e+C(3,1,0)-3-O(1,0,0)e-

N(2,0,0)-3-C(1,0,0)d+C(3,1,0)-3-C(2,2,1)d-

C(2,2,1)-3-N(2,0,0)c+C(3,1,0)-3-C(2,2,1)c-

C(2,2,1)-3-N(2,0,0)b+C(3,2,1)-3-C(1,0,0)b-

C(2,2,1)-3-N(2,0,0)a+C(3,2,1)-3-N(2,1,0)a-

C(3,1,0)-2(2)-O(1,1,0)3+N(2,1,0)-2(1)-O(1,0,0)3-

C(3,1,0)-2(1)-N(2,0,0)2+C(3,1,0)-2(2)-N(2,1,0)2-

C(3,2,1)-2(1)-N(2,0,0)1+C(3,2,1)-2(1)-C(3,1,0)1-

Positive APsNegative APs

2. Adding positive atom pairs to the fragment

X element type

n number of bonds to heavy atoms

p number of π bonds

r number of ring memberships

BO bond order

Atom Pairs 2 (AP2) : X1(n, p, r)-2(BO)-X2(n, p, r)

PAGE

How well does it work?

18

Organic Chemistry Database

Hristozov, D., Bodkin, M., Chen, B., Patel, H. & Gillet, V. J. 2011. Validation of Reaction Vectors for De Novo Design. Library Design, Search Methods, and Applications

of Fragment-Based Drug Design. American Chemical Society.

Reaction TypeNumber of

Reactions

Correctly Reproduced

Number Per cent

Epoxide reduction 450 449 99.8

Epoxide formation 450 444 98.7

Ester to amide 172 172 100.0

Alcohol dehydration 171 169 98.8

Claisen rearrangement 61 54 88.5

Beckmann rearrangement 123 123 100.0

Friedyl Crafts acylation 113 113 100.0

Olefin metathesis 9 7 77.8

Dieckmann condensation 98 91 92.9

Nitro reduction 231 230 99.6

Alkene oxidation 272 272 100.0

Cope rearrangement 453 306 67.5

Aldol condensation 134 134 100.0

Alcohol amination 97 97 100.0

Amide reduction 51 51 100.0

Diels-Alder hetero 441 320 72.6

Ether halogenation 58 58 100.0

Ozonolysis 132 125 94.7

Claisen condensation 98 77 78.6

Carboxylic acids to aldehydes 194 194 100.0

Nitrile reduction 102 102 100.0

Diels-Alder cycloaddition 106 65 61.3

Fischer indole 230 94 40.9

Alkene halogenation 310 281 90.6

Nitrile hyrdrolysis 460 460 100.0

Olefination 455 427 93.8

Wittig-Horner 211 190 90.0

Robinson annulation 13 10 76.9

Total 5,695 5,115 89.8

• ~3 seconds per reaction average,

0.015 seconds median run time

• Products generated for 5,115

reactions (~90% of the 5,695)

PAGE

Evolutionary Design

19

From fragment to drug-like molecules with designed polypharmacology

1. Multi-dimensional de novo design of drug-like compounds. 2013, De Novo Molecular Design ISBN 978-3-527-67700-9.

2. Validation of Reaction Vectors for de Novo Design. 2011, Library Design, Search Methods, and Applications of Fragment-Based Drug Design. ISBN 9780841224926.

3. Knowledge-Based Approach to de-Novo Design using Reaction Vectors. 2009, 49 (5), pp 1163–1184. J. Chem. Inf. Model.

Evotec Reaction

Transformation Database

Pool of possible products

In-silico reaction

Q2 = 0.76

Pre

dic

ted

Actual

Activity XQSAR

Local Models

Docking Score

Actual

Pre

dic

ted Q2 = 0.68

ADMETQSAR

Global Models

Multi-objective

Pareto optimisation

algorithm

Reaction Vectors Score Select

1. Using known chemistry for in-silico design and optimisation

Starting material

2. Evolutionary Algorithm Cream-off top scoring

molecules for evolution

• Only the best scoring molecules survive

and are resubmitted to the design

algorithm to genetically evolve into

better scoring solutions.

• Algorithm is well

suited to fragment

based design (FBDD)

approach wherein

small fragments are

grown into a protein

active site

• Methods to score the designs include: QSARs,

pharmacophore fit, docking scores, similarity to

known actives, polypharmacology prediction,

physical property predictions etc.

PAGE

Multi-Objective Optimisation

20

Haloperidol

Ziprasidone

“Score” = Similarity + D2predAct + a1BApredAct + H1predAct

Q2 = 0.76

Pre

dic

ted

ActualActualP

redic

ted

Q2 = 0.68

Actual

Q2 = 0.40

Pre

dic

ted

Union of Descriptor

+

PAGE

Results: Piperidone

21

26K Reactions, 93K Reagents

Starting Material

Fluphenazine

Chlorpromazine

PAGE

It looks great but ...

22

1. The algorithm only knows about transformation types that are in the Db!

2. The AP2/3’s cover 1 and 2 bonds. Remote functionality isn’t considered.

3. A reaction path is not a drug “optimisation”!

a b c d e

i ii iii iv

But how do we get from

here to here?

PAGE

Reaction Sequence Vectors

Tools for molecular design

xb

a b c d e

i ii iii iv

ea

a b

ca

da

Sequence Vectors

Molecules Nodes // RVs Edges

PAGE

SAR Exploration

24

Succinyl Hydroxamates

Bailey, S., et al., 2008. Bioorganic & Medicinal Chemistry Letters, 18, 6562-6567

PAGE

Novel SAR

25

Succinyl Hydroxamates

James Wallace ukQSAR 2014

PAGE

Principle Components Analysis of Property Space

26

Succinyl Hydroxamates

James Wallace ukQSAR 2014.

PAGE

Mapping Discovery Space

27

Sequence vector network

600K+ reactions from US Patent Database: Nextmove

PAGE

The PGC GWAS

28

Genome Wide Association Studies

PAGE

GPCRs associated with Schizophrenia GWAS genes

29

2 step shortest paths network

Suzanne Brewerton – UKQSAR Lilly 2014

Histamine receptor network

Adrenergic receptor network

Dopamine receptor network

Schizophrenia GWAS genes

GPCR receptor subtypes

Eg, From: Dopamine receptor subtypesTo: Proteins defined by PGC2 schizophrenia GWAS genes

Hubs involved in GPCR

signalling in schizophrenia

Remove proteins degree <10

PAGE

In-Silico Network Based Design

30

Protein Target

Deconvolution

Connecting Gene Expression Data from Connectivity Map and In Silico Target Predictions For Small Molecule Mechanism-of-Action Analysis

Bender et al., Molecular BioSystems 2015,11, 86-96

PAGE

Networks in Molecular Discovery

31

PhenoTarget - Phenotypic screening to target de-convolution

Focused set Chemically

diverse compounds

Biologically diverse

(annotated) compounds

Structurally

diverse hits

Hits with

target and

pathway

annotations

Selectivity

assays(Cerep, Kinaffinity)

Secondary

assays

Orthogonal

assays

Predictive

Pharmacology

Hit expansion

by structure

Structural

clustering

Hit expansion

by

target/pathway

Biomarker

development

Cellular Target

Profiling

Mechanism-

informed

phenotypic assay

Knockdown

studies

Target specific

assays

Gene targeting

mouse model

Compound

selection

Phenotypic

Screen

Build SAR Hit val.-H2L-LO In vivo studies

Biomarker

analysis PK/PD and MoA

Disease models

in vivo efficacy

Proteome / Transcriptome analysis

Evolve targets and pathways

Target

identification

and validation

App for both

routes

Target-gnostic

route and App’s

Target-agnostic

route and App’slegend

PAGE

An informatics workflow

32

When more is not always more

Target count Pathway count

Total Significant Total Significant

(1) Exact match 895 30 429 146

(2) + similarity

search

940 41 434 154

(3) + predictive

models

1380 134 471 226

PAGE

The Drug Discovery Challenge

33

Summary

Given a disease signature how do we best sample appropriate chemistry space?

PAGE

Happiness

34

What makes you happy?

A man doesn’t know what happiness is until he’s married,

by then it’s too late! (Frank Skinner)

PAGE

Acknowledgments

35

Dimitar Hristozov

David Thorner

David Evans

Nik Fechner

George Papadatos

Suzanne Brewerton

Lewis Vidler

Hina Patel

Ben Allen

James Wallace

John Liebeschuetz

Jason Cole

Colin Groom

Val Gillet

Beining Chen Andreas Bender

Georgios Drakakis

Tim James

Inaki Morai

Jon Hollick

York Rudhard

Alex Böcker

Your contact:

Building innovative

drug discovery alliances

Mike BodkinVP Research Informatics114 Innovation Drive,Milton Park, AbingdonOxfordshire OX14 4RZ, UKT: +44 (0)1235 44 1207

Mike.Bodkin@evotec.com

PAGE

Track record

37

Evotec Phenotypic DD projects and applied methods for TI/TV

1) Target ID achieved by known/annotated drug application

2) Annotated Cpd X prevents pericyte to myofibroblast transition and attenuates kidney fibrosis in vivo

Assay DevHit ID, Hit

validation

In vivo

studiesTarget ID

H2L LO

planned

B cell proliferate

Lung Fibrosis Pred.Pharm.diverse / bioannotated cpds

CTPT’omicsdiverse / bioann. cpds Neuroinflammation

Orphan transporter diverse cpds

Melanocytes Pred.Pharm.diverse cpds

Cancer immunotherapy CTPdiverse / bioann. cpds Pred.Pharm.

Diabetic nephropathy CTPT’omicsdiverse / bioann. cpds

Cancer double strand break repair CTPdiverse / bioannotated cpds Pred.Pharm.

diverse / bioann. cpds in vivo val Kidney fibrosis SimSearch2)

in vivo val Neuroprotection 1)

bioann. cpds SimSearch

Anti-viral anti-viral focussed library from Novira

Muscular dystrophy

Motor Neuron Disease

Others…

Target ID

In vivo val.

‘Omics

PAGE

Evolve Target & Pathway Hypothesis

Informing the Discovery Pipeline

38

Cheminformatics and Bioinformatics toolkit

Target Identification Ligand Identification Ligand Optimisation

Acquire the tools Evidence of SAR Explore the SAR envelope

Phenotypic

ScreenBuild SAR Hit Val-H2L-LO

Target Identification

& Validation

Compound

Selection

2. Target Unknown

1. Target Known

• Chemically

diverse

• Biologically

diverse

• Different operational paradigms apply depending on whether the target is known