Chris de Graaf – Director, Head of Computational Chemistry

Integrating Artificial Intelligence Approaches to Enhance GPCR Drug Discovery

NON-CONFIDENTIAL

12 September 2019 | R&D Investor Day

2

Chris de Graaf – Director, Head of Computational ChemistryAbout Me

DUTCH HERITAGE;UK-BASED

UNIVERSITY OF AMSTERDAM

MSc.

VRIJE UNIVERSITEIT AMSTERDAM

PhD.

POSTDOCTORAL RESEARCH FELLOW

3 YEARS

DIRECTOR, COMPUTATIONAL

CHEMISTRY2 YEARS

ASSOCIATE PROFESSOR

10 YEARS

3

Introduction

Chief R&D Officer

Platform Technology Drug Discovery Preclinical

DevelopmentClinical Drug Development

Comp.Chemistry

Medicinal Chemistry

Molecular Pharmacology

Clinical DevelopmentProtein Engineering

Biomolecular Structure

Biophysics

Translational Sciences

4

What is Computational Chemistry and why is it important for Drug Discovery?

By harnessing the power of computers and data, Computational Chemistry strives to make the drug discovery more efficient

Integrates chemical theory and modelling with experimental observations using algorithms, statistics and large databases

Simulates physical processes and uses statistics/data analyses to extract useful information from large bodies of data

Presents complex analyses in an understandable form to design experiments and new materials and validate results

Influences our understanding of the way the world works and characterizes new compounds and materials

Important in pharma industry to discover and design new therapeutics and apply cheminformatics to processes data

Sosei Heptares’ GPCR structures have changed the face of drug design

Sources: Langmead et al. J. Med. Chem. 2012, 1904; Congreve et al. J. Med. Chem. 2012, 1898

The Computational Chemistry Field is enabling the new era of GPCR Structure-Based Drug Design

• Detailed structural insights into GPCR binding sites enable atom by atom design and optimization of GPCR-drug interactions using computational chemistry approaches

• Efficient optimization of physiochemical properties of drug molecules improves their pharmacokinetics, efficacy, and safety

5

Structure-based Virtual Screen

(VS)

• Poor physiochemical properties• Furan containing

• Novel non-furan containing• Moderate selectivity (vs A1)

Hit 1Preladenant

NH2

NN

S

N

OH

O

N

N

N

N

O

NN

NH2NN

O

NH2

N

N N

N

F

Cl

AZD4635

BPM & VSSBDD

Design vector

SBDDLipophilic

hotspots water networks• Novel triazene template• No structural alerts• Low selectivity (vs A1)• Mod. metabolic selectivity

NH2

N

N N

Hit 2

• Improved LLE• Improved selectivity• Improved metabolic stability

ARTIFICIAL INTELLIGENCECapability of machines to perform a task by

learning from data

Artificial Intelligence (AI) and Machine Learning (ML) in Drug Discovery

AI has the potential to supercharge computational chemistry and cheminformatics approaches for GPCR structure-based drug discovery using Sosei Heptares’ integrated GPCR databases

6

MACHINE LEARNINGAI technique to automatically identify

relationships between input and output data

NEURAL NETWORK

Machine Learning algorithm consisting of different layers connecting input and output

data

Solving well-defined

problems using rich

training data

Drug discovery

Solving a dynamic,

multi-objective problem? Sparse chemical, biological,

structural dataKnowing how drugs

bind in 3D

Artificial Intelligence (AI) and Machine Learning (ML) in GPCR SBDD

Sosei Heptares proprietary GPCR databases give us a competitive advantage in the identification of novel patterns and relationships between GPCR biological, chemical and structural data

7

PAR2Cheng et al., Nature (2017)

CCR9Oswald et al., Nature (2016)

GCGRJazayeri et al., Nature (2016)

CRF1Hollenstein et al.,

Nature (2013)

C5aRobertson et al.,

Nature (2018)

GPCR structures GPCR binding mode diversityA B

Extending the Chemical Space of the Structural GPCRome

Source: 1 Protein Data Bank. Includes 30 GPCR complexes from Sosei Heptares disclosed in the public domain

Sosei Heptares proprietary structures enable us to uniquely extend the bioactive chemical space for GPCRs and design new drugs and explore new modes of GPCR modulation

8

A

GLP-1R GCGR

CT

PTH1R

CALCRCRF1

mGlu5mGlu1

FZD1

SMO

CCR5CCR2

CXCR4

C5a1BLT1

APJAT1 AT2

DP2

δOR

µORκORNOP

LPA6

P2Y1

FFA1P2Y12

PAR1PAR2

PAF

TA2R

EP4

EP3

D4D3

D2M1

M3M2M4

5HT1B

5HT2B

5HT2C

H1

5HT2A

β1β2

Rho

MT1

MT2

A1A2A

CB1

CB2

S1P1

LPA1ETB

Ox1

Ox2

NK1

NPYY1

Structural GPCRome: GPCR-ligand structures

211Structures GPCR-ligand complexes

publicly available1

230+Unreleased structures GPCR-ligand

complexes Sosei Heptares

60+GPCR StaRs

Chemical coverage of the Structural GPCRome

chemical similarity

790.383 GPCR-ligand complexes (public)

114.673 GPCR-ligand complexes (similar)

Extending the Binding Site View on GPCR LigandsB

9

Sosei Heptares combines structural information – including key structures only known to us –with cutting edge computational methods to identify, compare and analyze GPCR binding sites for SBDD

Chemical coverage of the Structural GPCRomeStructural GPCR-Ligand Interaction Descriptors for Computer-Aided Drug Design

lipophilic hotspots

water networks

similarity

cryptic pockets

GPCRstructural

chemogenomics

10

How can we realistically deploy AI at Sosei Heptares?

Drug Discovery Preclinical Development

Clinical Drug Development

Commercial,Life cycle

• Validated target• Assay developed• Lead molecule demonstrated

effect on target

Out

com

esAI

App

licat

ion

• Prediction of role in target in disease

• Identication novel hit molecule (Virtual Screening)

• Design compound libraries• Prediction of drug-target

interaction• Prediction of drug impact on

signalling pathways

• Lead candidate demonstrated effect in preclinical model

• Prediction of ADMET (Absorption, Distribution, Metabolism, Excretion, Toxicity) properties

• Drug demonstrates:• Safety in humans• Efficacy in humans• Efficacy in large patient

population

• Drug repurposing• Selection of patient population• Review of patient data

(e.g. genomic data)

• Marketing authorization• Commercial manufacture• Post-marketing surveillance• Adverse effect

• Pharmacovigilance

Source: Adapted from Biopharma Excellence

AI approaches will focus on key areas in R&D that using unique Sosei Heptares GPCR data

11

We are actively rolling out AI across our R&D platform

Platform Technology Drug Discovery Preclinical

DevelopmentClinical Drug Development

Medicinal Chemistry

Molecular Pharmacology

Clinical DevelopmentProtein Engineering

Biomolecular Structure

Biophysics

Computational Chemistry Translational Sciences

Artificial Intelligence for Multi-Parametric GPCR Drug Discovery

Machine Learning

Data & descriptors

AI StaR® Design1

Ligand Design2

ADMET Prediction3 Target / Indication

Selection4

Sosei Heptares AI Platform – StaR® Design

StaR® Design

sequence alignment analysis structural signatures

5-HT1B 2 2 2 2 2 2

5-HT2B 4 5 5 5 5

5-HT2C 1 1 1 1

β1 2 6 5 30 9 28 34 32 34

β2 2 3 1 1 17 4 14 18 18 18

D2 1 1 1 1 1 1 1 1

D3 2 2 1 2 2 2

D4 2 2 2 2 2 2 2 2 2 2 2

H1 2 2 2

M1 1

M2 1 1 1 1 3 1 3

M3 1 9

M4 2

AT1 1 2 1 2 2 2 2 2 1 2 2

AT2 5 2 3 5 5 5 3 5 2 5 5

C5a1

ETB 1 2 1 1 1 3 1 2 1 1 3 3 3 3

NTS1 10 8 8

δ 2 4 6 6 6

κ 2 2 2 2 2 2 2 2 2 2

µ 1 1 1 1 2 2 2

NOP 6 4 6 2 6 1 5 4 6 6 6

OX1 2 2 2 2 2 1 2 2 2

OX2 1 1 1 1 1 1 1

PAR1PAR2 1 1 1 1 1 1 1 1 1 1 1 1 1 1

CCR2 1 1 1 1 1 1 1 1 1 1 1

CCR5 1 1 2 1 2 2 4 1 4 2 1 1 3 2

CXCR4 1 10 6 10 10 10 7 10 9 10 9

US28 2 1 2 2 2 2 2 1 2 2 2 2

BLT1 1 1 1 1 1 1 1 1 1 1

CB1 2 2 2 4 3 2 4 2 2 1 4 4

FFA1 3 3 3

LPA1 2 3 3 3 1 2 2 3 3 3

S1P1 2 1 1 2 1 2 2 2 2

A1 2 2 3

A2A 2 5 1 5 6 2 1 11

P2Y1 1 1 1 1

P2Y12 2 2 2 2 2 3

TM33.30 3.32

Chem

okine

Lipid

Nucle

otide

3.26 3.27 3.28 3.2923.50 3.21 3.25IFP

Amine

rgic

Pepti

de

2.59 2.60 2.61 2.63ECL1

Receptor 1.27 1.30 1.31 1.32 1.34 2.64 2.651.35 1.39 2.53 2.54 2.56 2.57TM1 TM2

structural interaction fingerprints

Mining unique proprietary information on structural determinants of GPCR thermostabilisation to accelerate GPCR structure determination for SBDD

1

GPCR structure

Biomolecular StructureProtein

Engineering

Bioinformatics StaR data

Machine Learning

12

Sosei Heptares AI Platform – Ligand Design

Ligand Design

Interaction Fingerprint (IFP) based virtual GPCR ligand screening

Identifying GPCR specific structural interaction features that enable efficient Virtual Screening for novel hit molecules

2

Computational Chemistry

Cheminformatics

Medicinal Chemistry

GPCR SBDD

Machine Learning

5-HT1B 2 2 2 2 2 2

5-HT2B 4 5 5 5 5

5-HT2C 1 1 1 1

β1 2 6 5 30 9 28 34 32 34

β2 2 3 1 1 17 4 14 18 18 18

D2 1 1 1 1 1 1 1 1

D3 2 2 1 2 2 2

D4 2 2 2 2 2 2 2 2 2 2 2

H1 2 2 2

M1 1

M2 1 1 1 1 3 1 3

M3 1 9

M4 2

AT1 1 2 1 2 2 2 2 2 1 2 2

AT2 5 2 3 5 5 5 3 5 2 5 5

C5a1

ETB 1 2 1 1 1 3 1 2 1 1 3 3 3 3

NTS1 10 8 8

δ 2 4 6 6 6

κ 2 2 2 2 2 2 2 2 2 2

µ 1 1 1 1 2 2 2

NOP 6 4 6 2 6 1 5 4 6 6 6

OX1 2 2 2 2 2 1 2 2 2

OX2 1 1 1 1 1 1 1

PAR1PAR2 1 1 1 1 1 1 1 1 1 1 1 1 1 1

CCR2 1 1 1 1 1 1 1 1 1 1 1

CCR5 1 1 2 1 2 2 4 1 4 2 1 1 3 2

CXCR4 1 10 6 10 10 10 7 10 9 10 9

US28 2 1 2 2 2 2 2 1 2 2 2 2

BLT1 1 1 1 1 1 1 1 1 1 1

CB1 2 2 2 4 3 2 4 2 2 1 4 4

FFA1 3 3 3

LPA1 2 3 3 3 1 2 2 3 3 3

S1P1 2 1 1 2 1 2 2 2 2

A1 2 2 3

A2A 2 5 1 5 6 2 1 11

P2Y1 1 1 1 1

P2Y12 2 2 2 2 2 3

TM33.30 3.32

Ch

em

okin

eL

ipid

Nu

cle

oti

de

3.26 3.27 3.28 3.2923.50 3.21 3.25IFP

Am

ine

rg

icP

ep

tid

e

2.59 2.60 2.61 2.63ECL1

Receptor 1.27 1.30 1.31 1.32 1.34 2.64 2.651.35 1.39 2.53 2.54 2.56 2.57TM1 TM2

IFPs

13

Sosei Heptares AI Platform – Ligand Design

Ligand Design AI molecule generation models and scoring approaches using proprietary structural descriptors and customized to GPCR targets

2

Computational Chemistry

Cheminformatics

Medicinal Chemistry

GPCR SBDD

Reinforcement Learning

Artificial Intelligence driven molecule generation

Recurrent Neural Networkscoring

14

Sosei Heptares AI Platform – Ligand Design

Ligand Design

3D alignment/docking

2

Computational Chemistry

Cheminformatics

Medicinal Chemistry

GPCR SBDD

iteration GPCR

-liga

nd d

ocki

ng sc

ore

3D li

gand

sim

ilarit

y sc

ore

GPCR-ligand docking score3D li

gand

sim

ilarit

y sc

ore

iteration

Multiparameter scoring using proprietary structural descriptors

Artificial intelligence driven molecule generation

AI molecule generation models and scoring approaches using proprietary structural descriptors and customized to GPCR targets

15

Sosei Heptares AI Platform – Ligand Design2

Representative ligand LHS and variable RHS

FEP based ligand binding mode prediction

The computational prediction of ligand binding affinity has been a treasured goal for many years: FEP has been customised for GPCR SBDD through a collaborative arrangement with the leaders in the field

Free Energy Pertubation (FEP) In-house GPCR structures FEP vs. experimentally determined ligand affinity

16

Sosei Heptares AI Platform – ADMET Prediction

Ligand Design Combining external/in-house data sets to train ADMET prediction models to guide the design of GPCR ligands with improved drug properties

3

Computational Chemistry

Cheminformatics

Medicinal Chemistry

GPCR SBDD

ADMET Prediction

Hit 2A2A pKi 6.9

Mod solubilityLow selectivity (vs A1)

Mod.metabolic stability

HTL1071/AZD4635A2A pKi 8.0

Improved solubilityImproved selectivityImproved metabolic

stability

SBDD

A1A2A

NH2

N

N N

N

F

ClNH2

N

N N

Solubility prediction model

Hepatic Clearance prediction model

17

Sosei Heptares AI Platform – Target / Indication Selection4

Target / Indication Selection

Biology/GenomicsPharmacology

TranslationDevelopment

GPCR structureBioinformatics Identification Disease Associated GPCR Variants for SBDD

Genome/phenome data mining to identify

GPCR variants associated with

specific disease areas

PHEnomeRare Diseases

VariantsUK Biobank 100K genomes gnomADCOSMICClinVar

GPCRomeSelection StAR design

AgonistAntagonist

CHEMomeDesign

In Vitro ScreenCombinationsMachine Learning

Structural cheminformatics

analysis GPCR variants

Comparative pharmacological

evaluation identified GPCR variants

Generation StaR of disease associated GPCR variant for SBDD

18

Uniquely positioned to leverage next generation AI / ML methods

FASTERStaR® generation

1

BETTERGPCR ligand design

2

PREDICTEDADMET properties

3

SMARTERGPCR target /

indication selection

4

Artificial Intelligence (AI) and Machine Learning (ML) can be gamechangers for drug discovery if used with the correct understandingof how drug ligands bind

Sosei Heptares has the technology to gather unique insights in thecomplex biological, structural and chemical space of GPCRs andtheir ligands

Sosei Heptares’ proprietary engine delivers innovative new GPCRstructures, enabling the generation of GPCR SBDD data andtechnologies that uniquely enable AI / ML

Sosei Heptares’ proprietary GPCR databases, in-house knowledge,and collaborative initiatives enable AI / ML driven data mining thatgive us a competitive advantage in GPCR drug discovery

19

20

The material that follows is a presentation of general background information about Sosei Group Corporation and its subsidiaries (collectively, the “Company”) as of the date of this presentation. This material has beenprepared solely for informational purposes and is not to be construed as a solicitation or an offer to buy or sell any securities and should not be treated as giving investment advice to recipients. It is not targeted to the specificinvestment objectives, financial situation or particular needs of any recipient. It is not intended to provide the basis for any third party evaluation of any securities or any offering of them and should not be considered as arecommendation that any recipient should subscribe for or purchase any securities.

The information contained herein is in summary form and does not purport to be complete. Certain information has been obtained from public sources. No representation or warranty, either express or implied, by theCompany is made as to the accuracy, fairness, or completeness of the information presented herein and no reliance should be placed on the accuracy, fairness, or completeness of such information. The Company takes noresponsibility or liability to update the contents of this presentation in the light of new information and/or future events. In addition, the Company may alter, modify or otherwise change in any manner the contents of thispresentation, in its own discretion without the obligation to notify any person of such revision or changes.

This presentation contains “forward-looking statements,” as that term is defined in Section 27A of the U.S. Securities Act of 1933, as amended, and Section 21E of the U.S. Securities Exchange Act of 1934, as amended. Thewords “believe”, “expect”, “anticipate”, “intend”, “plan”, “seeks”, “estimates”, “will” and “may” and similar expressions identify forward looking statements. All statements other than statements of historical facts included inthis presentation, including, without limitation, those regarding our financial position, business strategy, plans and objectives of management for future operations (including development plans and objectives relating to ourproducts), are forward looking statements. Such forward looking statements involve known and unknown risks, uncertainties and other factors which may cause our actual results, performance or achievements to bematerially different from any future results, performance or achievements expressed or implied by such forward looking statements. Such forward looking statements are based on numerous assumptions regarding ourpresent and future business strategies and the environment in which we will operate in the future. The important factors that could cause our actual results, performance or achievements to differ materially from those in theforward looking statements include, among others, risks associated with product discovery and development, uncertainties related to the outcome of clinical trials, slower than expected rates of patient recruitment,unforeseen safety issues resulting from the administration of our products in patients, uncertainties related to product manufacturing, the lack of market acceptance of our products, our inability to manage growth, thecompetitive environment in relation to our business area and markets, our inability to attract and retain suitably qualified personnel, the unenforceability or lack of protection of our patents and proprietary rights, ourrelationships with affiliated entities, changes and developments in technology which may render our products obsolete, and other factors. These factors include, without limitation, those discussed in our public reports filedwith the Tokyo Stock Exchange and the Financial Services Agency of Japan. Although the Company believes that the expectations and assumptions reflected in the forward-looking statements are reasonably based oninformation currently available to the Company's management, certain forward looking statements are based upon assumptions of future events which may not prove to be accurate. The forward looking statements in thisdocument speak only as at the date of this presentation and the company does not assume any obligations to update or revise any of these forward statements, even if new information becomes available in the future.

This presentation does not constitute an offer, or invitation, or solicitation of an offer, to subscribe for or purchase any securities. Neither this presentation nor anything contained herein shall form the basis of any contract orcommitment whatsoever. Recipients of this presentation are not to construe the contents of this summary as legal, tax or investment advice and recipients should consult their own advisors in this regard.

This presentation and its contents are proprietary confidential information and may not be reproduced, published or otherwise disseminated in whole or in part without the Company’s prior written consent. These materialsare not intended for distribution to, or use by, any person or entity in any jurisdiction or country where such distribution or use would be contrary to local law or regulation.

This presentation contains non-GAAP financial measures. The non-GAAP financial measures contained in this presentation are not measures of financial performance calculated in accordance with IFRS and should not beconsidered as replacements or alternatives profit, or operating profit, as an indicator of operating performance or as replacements or alternatives to cash flow provided by operating activities or as a measure of liquidity (ineach case, as determined in accordance with IFRS). Non-GAAP financial measures should be viewed in addition to, and not as a substitute for, analysis of the Company's results reported in accordance with IFRS.

References to "FY" in this presentation for periods prior to 1 January 2018 are to the 12-month periods commencing in each case on April 1 of the year indicated and ending on March 31 of the following year, and the 9 monthperiod from April 1 2017 to December 31 2017. From January 1 2018 the Company changed its fiscal year to the 12-month period commencing in each case on January 1. References to "FY" in this presentation should beconstrued accordingly.

Sosei Heptares is a trading name. Sosei and the logo are Trade Marks of Sosei Group Corporation, Heptares is a Trade Mark of Heptares Therapeutics Limited. StaR is a Trade Mark of Heptares Therapeutics Limited

Disclaimer

Thank you for your attention

SOSEI HEPTARES

PMO Hanzomon 11F

2-1 Kojimachi, Chiyoda-ku

Tokyo 102-0083

Japan

Steinmetz Building

Granta Park, Cambridge

CB21 6DG

United Kingdom

North West House

119 Marylebone Road

London NW1 5PU

United Kingdom