MDC Connect: In-Silico Drug Design - what to do, what not to do - project driven examples

Exploiting medicinal chemistry knowledge to accelerate projects May 2020
May 2020
Not for Circulation
In-Silico Drug Design
What to do, What not to do
Project driven examples
Available on Slideshere - search for Dossetter
Twitter @MedChemica
Twitter @covid_moonshot
Twitter #BucketListPapers
www.medchemica.com/bucket-list/

Exploiting medicinal chemistry knowledge to accelerate projects May 2020Exploiting medicinal chemistry knowledge to accelerate projects May 2020
Not for circulation
About Me!
• I am an organic chemist by training:-
– Ph.D. University of Nottingham (Steve Clark) Jan 1998.
– Post-Doctoral - Harvard University – Eric Jacobsen
• I am a medicinal chemist by choice:-
– AstraZeneca 13 years
– Worked on Oncology, Inflammation, Diabetes and Obesity.
– Enzymes, GPCRs, nuc-hormone receptors, CNS targets
• Started MedChemica in 2012
– Taking Matched Molecular Pair Analysis as an AI technique to the next
level…
• Let me focus on sharing some of the lessons I have learnt…
– These are my own experiences from >20 years in pharma.

Not for circulation
Agenda
• Problem statement
– About me!
– Too much to talk about in 50 mins!
– What is the reason to use In-Silicon Drug Design?
• Where do Ideas come from?
• Chem-infomatics
– Get as much information as you can (literature / patents / structures)
• 2D Design - Prediction methods
– Understanding how “models” are built what their limitations are.
– Methods – QSAR / Machine Learning / MMPA and AI
• -> using AlogP98 example
• 3D Design / Structure Based Drug Design (SBDD)
– Methods – 3D modelling, docking what are their limitations
• Two project examples
– Cathepsin K – today’s example
– 11-bHSD – ask for a longer presentation
• What to do / What not to do - Conclusions

Exploiting medicinal chemistry knowledge to accelerate projects May 2020May 2020
Not for Circulation
The Problem Statement?
What is the reason to use In-Silicon Drug Design

Not for circulation
…8 Years of working with pharma companies
“Our median number of compounds per LO project is 3000 - this is
unsustainable… [it should be] 300” – Director of Chemistry (large pharma)
“Can we define the text book of medincal chemistry?”
– Director of Comp Chem (large pharma)
“We are aiming at 300 compound per project – currently we are about
400, we will get better” – ExScienta scientist at SCI ‘What can BigData do
for chemistry’ – Oct 2017
MedChemica is a company using knowledge extraction techniques to
build “expert systems” to suggest actions to chemists [Artificial
Intelligence – AI]
We use In-Silico techniques to analyse rigourously and refine our
compound designs to achieve the goal of a quality candidate drug in
fewer iterations.
‘Diagnosing the decline in pharmaceutical R&D efficiency’ - Scanel, Blanckley, Boldon,
Warrington, Nature Reviews Drug Disco., 2012, 11, 191
Hopkins, A.L.; Mason, J.S.; Overington, J.P. Can we rationally design promiscuous drugs?
Curr.Opin.Struct.Biol, 2006, 16, 127.

Not for Circulation
Where do Ideas come from?
Chem-infomatics
…so what are you going to make next…?
Get as much information as you can!
Data from the literature and patents
Free-Wilson analysis

In-Silico Drug Discovery
• The only way to Make and Test fewer compounds is better Design.
• Without quality analysis any designs will be poor.
• In-Silico Drug Discovery covers both Analysis and Design.
Design
In-Silico
Drug Discovery
Analyse Make
Test

References DTMA cycles
Sewing, A Drug Disco. Techno, 2009, DOI, 10.1016/j.ddtec,2008,12.002
Andersson S et al, Making medicinal chemistry more effective--application of Lean Sigma to improve processes, speed
and quality. Drug Discov Today. 2009 Jun;14(11-12):598-604.
Johnstone, C.; Pairaudeau, G.;Pettersson, J. A.; Creativity, innovation and lean sigma: a controversial combination? Drug
Discov Today. 2011 Jan;16(1-2):50-7
Robb, G.R.; McKerrecher, D.;Newcombe, N.J.;Waring, M.J. A chemistry wiki to facilitate and enhance compound design in
drug discovery. Drug Discov Today. 2013 Feb;18(3-4):141-7.
Plowright, A.T.; Johnstone, C.; Kihlberg, J.; Pettersson, J.; Robb, G.; Thompson, R.A.; Hypothesis driven drug design:
improving quality and effectiveness of the design-make-test-analyse cycle. Drug Discov Today. 2012 Jan;17(1-2):56-62
Baldwin, E.T., Metrics and the effective computational scientist: process, quality and communication. Drug Discov Today.
2012 Sep;17(17-18):935-41.
Cumming, J.G.; Winter, J.P.; Poirrette, A. Better compounds faster: the development and exploitation of a desktop
predictive chemistry toolkit. Drug Discov Today. 2012 Sep;17(17-18):923-7.
Baede, E.J.; Bekker, E.J.W.; Cronin, D.;Integrated project views: decision support platform for drug discovery project teams.
J Chem Inf Model. 2012 Jun 25;52(6):1438-49.
Contrast to:-
MacDonald, J. F.; Smith, P. W. Lead Optimization in 12 months? True confession of a chemistry Team Drug Discovery Today,
2001, 6, 18, 947
The best results were achieved by encouraging team work and increasing CLARITY through
effective COMMUNICATION

Which molecules to make next?
Often Med Chemists stick to what they can make..
In-Silico design helps us process the volume of data and make better decisions
Lead Generation
targeted libraries
‘lead-like’ & ‘drug-like’
molecules
Physicochemical
Properties
log P, pKa,
H-bonding, solubility
Computational
Chemistry
3-D molecular properties
receptor & enzyme models
QSAR,chemoinformatics
Biological Properties
human molecular target
receptor families
in vitro affinity & efficacy
selectivity & toxicity
in vivo disease models
Patents and
Publications
competition
SAR
Synthesis
traditional, parallel
combinatorial
Medicinal Chemist’s Toolbox
Metabolism &
Pharmacokinetics
clearance, metabolism,
oral bioavailability,
duration
Volumes of Data
Analysis tools,
Visualisations,
Peer support groups
Design Teams
Targets and Goals
Meetings and Time Pressure
Competition
Internal,
External
New Technology
Latest thing X
Innovation demands
TIME

Not for circulation
Free Wilson Analysis – SAR fragments
Free, S.M.; Wilson, J.W.; A Mathematical Contribution to Structure-Activity Studies J. Med. Chem. 1964, 7, 4, 395-399
Gillet, V.J.; Willett, P.; et al Assessment of Additive/Nonadditive Effects in Structure−Activity Relationships:
Implications for Iterative Drug Design, J. Med. Chem. 2008, 51, 23, 7552-7562

Extracting data from the literature and
patents O
ON
O
N
N
HN Cl
F
O
O
O
O
N
N
HN
Graph analytics
community detection
Extracting as much info as
possible…

Recent GPCR target, some literature, some patents…..
Problem: what’s the SAR in this data set?
Pair
Finding
Rule Finding
Structures
&
Data
High potency
groups
Fragment LLE
“rules” for structural changes
that increase potency
Identify pivotal compounds
Pairs to the pivotal compound: potency
by ClogP = fragment LLE
Chemical changes
with large
DpIC50 /DlogP
443 compounds
In-Silico analysis is un-biased, rigorous and quicker -> better design

Not for Circulation
2D and 3D Design
What techniques?

Theoretical Approaches (selected)
2D Techniques (predicting properties of a new molecule)
• QSAR – Quantitative Structure Activity Relationships
• Group contribution/Hunter approach (AlogP98 example)
• Matched Molecular Pair Analysis (MMPA)
• Machine (Deep) Learning
3D Techniques
• Single Molecule structures
• Conformational and torsional analysis
• Protein / Ligand structures
• Docking / Scoring
• Force field / simulations Cresset
• FEP – Free Energy Perturbation Cresset
• QM – Quantum mechanical Andrew G. Leach
• MD – Molecular dynamics Pat Walters Relay Thera

It’s all lipophilicity?
Lipophilicity!
“Every time you use ClogP,
and make a decision, you are
doing In-Silico Drug Design”
However ClogP is not a
panacea for fixing all
problems, and it is not to be
trusted

Not for circulation
But the literature says it’s lipophilicity
Waring M. et al Bio. Org .Med. Chem .Lett, 2007, 15, 1759
Waring M. Expert Opin. in Drug Disco. 2010, 235.
‘The focus on Ro5 is oral absorption and the rule neither
quantifies the risk of failure associated with non-
compliance nor provides guidance as to how sub-
optimal characteristics of compliant compounds might
be improved’
Kenny, P. W.; Montanari, C. A. J. Comput Aided Mol Des,
2013, 27, 1-13.
Sweet spot in the middle

Reducing lipophilicity will solve the ADMET issue
Ph to 4-sub phenyl HLM
Slope of line 0.33
-0.4
-0.2
0
0.2
0.4
0.6
0.8
ΔLog10HLMClint
4-CH2CH3
4-CH3
4-CN
4-CONH2
4-COOH
4-F
4-NHSO2CH3
4-OCF3
4-OCH3
4-Ph4-SO2CH3
4-SO2NH2
-3 -2.5 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2
ΔLogD_pH7.4
Small group substituents on phenyl and HLM stability Dossetter, A. G. Bio-org. Med. Chem. 2010, 4405
P4503A4, hERG, PAMPA - Gleeson, P.; Bravi, G.; Modi, S.; Lowe, D. Bio-org. Med. Chem. 2009, 17, 5906
hERG / Log D
Slope of line 0.30
3A4 inhib / Log D
Slope of line 0.16
…thus a 10 fold change will require > 3 log unit drop

571 compounds from
799
are >10 µM
71%
102 compounds
from 185 are >10
µM
55%
513 compounds from
594
are >10 µM
86%
102 compounds
from 122
are >10 µM
83%
509 compounds
from 704
are >10 µM
72%
41 compounds
from 256 are >10
µM (14%)
86% <10 µM
237 compounds
from 329
are >10 µM
72%
60 compounds
from 163
are >10 µM
(37%)
63% <10 µM
MW 500
LogD
2.5
Inhibition of Cyp 3A4
Dependency on logD7.4 and MW in inhibition of 3A4
58% of compounds are >10 µM
Whole dataset
χ2 6.5 R2 0.0035 CMH χ2 11.6
Three years later, new compounds
through the same assay - the relationship
has disappeared
MW 500
Up to Nov 2003 (n = 1452) Nov 2003 to July 2006 (n = 1700)
Do NOT bin data to generate ’Rules’
2D plots only, and measured data on a continuous linear scale
Inflation of correlation in the pursuit of drug-likeness. Kenny, P.W. & Montanari, C.A. J Comput Aided Mol Des (2013) 27: 1.

Appreciating how QSAR models are built?
Wildman, S.A.;Crippen, G.M.; Prediction of Physicochemical Parameters by Atomic Contributions
J. Chem. Inf. Comput. Sci. 1999, 39, 868-873
Paper accounts for the environment by different types of atom
e.g. 27 types of carbon atom
All models are only as good as the dataset that produce them.
A new functional group, with a new type of C atom, could produce a different results.
Intramolecular H-bonds or ’masked’ polar groups also produce different results.

References on model building
Dearden, J.C.; Cronin, M.T.D.; Kaiser, K.L.E.;
How not to develop a quantitative structure–activity or structure–property
relationship (QSAR/QSPR).
SAR QSAR in Environ. Res. 2009, 20, 241-266.
Maggiore, G.M.;
On outliers and activity cliffs – why QSAR often disappoints.
J. Chem. Inf. Model. 2006, 46, 1535.

Not for Circulation
Matched Molecular Pair Analysis
How our Artificial Intelligence (AI) system works

Not for circulation
A B pSol A (μM) pSol B (μM) ∆pSol
- 4.1 (77 μM) - 3.1 (870 μM) 1.0
- 6.0 (1.0 μM) - 3.7 (178 μM) 2.3
-5.7 (2.0 μM) - 4.1 (82 μM) 1.6
3 pairs +ve Sol
Median 1.6
CHEMBL2325741CHEMBL2325742
From SAR to MMPA…..
CHEMBL3356658 CHEMBL218767
CHEMBL456322CHEMBL456802
MCPairs Rule finder required 6 matched pairs for 95% confidence

Not for circulation
From SAR to MMPA…..
Actual Rule from MCPairs
Endpoint:
Aqueous Solubility at pH 7.4
[CHEMBL2362975]
n-qual 69
n-qual-up 47
n-qual-down 21
median ∆pSol 0.26
std dev +/- 0.636

Not for circulation
“Multi-Step” transformations
Shibuya Crossing Tokyo
A C
B
E
F
Would you go steps via A -> B -> C
How would you go know to go E -> F
Or go straight there via D
- if the data said it was good?
D

Not for Circulation
Project example
Cathepsin K Project
Understanding the data you have
Minimising the scaffold
3D – Structure based design
CD1 and CD2 compounds
Crawford, J.J.; Dossetter, A.G J Med Chem. 2012, 55, 8827.
Dossetter, A.G. Bioorg. Med. Chem. 2010, 4405.
Dossetter, A.G. et al Bioorg Med Chem Lett. 2012, 22(17), 5563-5568.
Dossetter, A.G. et al J Med Chem. 2012, 55(14), 6363-6374.

Cathepsin K is a Serine Protease
AZ11961622
CTK2 / pIC50 8.96 (n=6)
CatS, CatL, CatB /pIC50 7.4 (n=6), 6.6, 7.0
hERG pIC50 <4.5
MW 492 Da
LogD (ClogP) 3.6 (4.3)
PPB Human (% Free) 5.1% (nd rat)
Solubility 99 µM
Clint Rat heps (hum mic) 20 (127) mL/min/kg
P450 (µM) 3A4 3.2 mM
F
N
N
O
NH O
N
O
Series closest to CD quality
Minimum requirements are mprovements in stability and selectivity
Reversible covalent binding to an electrophilic nitrile

Strip back and make simple compounds…..
Compound R amine Cat K
enzyme
pIC50
LE / LLE
AZ12586095 6.5 0.53 / >6.5
AZ12483590 7.1 0.49 / 6.3
AZ12594014 7.4 0.48 / 7.3
AZ12483591 8.2 0.43 / 6.4
AZ12603443 7.8 0.53 / 6.9
AZ12483607 8.8 0.50 / 6.7
N
N
N N
N N
N
O
N
O
N
R
N
Clear rigid SAR
Explained by x-tal
structure
Parent compounds
tend to have highest
inhibition
“Simple compounds
improve the in-silico
models”

Importance of electron density
Pike, A. C. W.; Hubbard, R. E. ; et al
Nature 1997, 389, 753 - 758.

AZ12483591
Cat K pIC50 8.1
Hu Mics Clint 6.1
Rat Mics Clint 115
AZ12475452
Cat K pIC50 7.2
Hu Mics Clint <2
Rat Mics Clint 11.9
Tempting to make 1200 compounds (secondary amines) but…..
Cyclohexyls – high confidence they bind close to conformation energy
minima. i.e. minimisation and docking close to real structures.
P3 ‘grove’ open and possible to optimise against a tyrosine and backbone
P3
Tyr

3D structure analysis checklist: • Check electron density / Check RMSE
• Take the small molecule, minimise and dock,
compare to xtal structure – how close?
• A QM calculation of 3D shape is worth the
investment here
• Check electro-static interactions
• Are the distances and angles good?
• Aligned to lone pairs
• Distance of 2.5 Angstroms
• Pi stacking, edge to Pi stacking – efficient
• Unlikely to optimise a hydrogen bonding
interaction – can be done but hard
• More likely to minimise the conformational energy of the
small molecule.
• Look for polar groups in un-favourable places
• Expand you molecule where there is space.
P3
Tyr “A Medicinal Chemist’s Guide to Molecular Interactions” Bissantz, Kuhn and Stahl
J. Med. Chem. 2010, 53, 5061-5084
“Intramolecular Hydrogen Bonding in Medicinal Chemistry” Kuhn, Mohr and Stahl
J. Med. Chem. 2010, 53, 2601–2611
“Application and Limitations of X-ray Crystallographic Data in Structure-Based Ligand
and Drug Design” Davis, A.M.;* Teague, S.J.; Kleywegt, G.J.; Angew. Chem. Int. Ed.
2003, 42, 2718 – 2736
“Hydrogen Bonding, Hydrophobic Interactions, and Failure of the Rigid Receptor
Hypothesis.” Davies, A.M.; Teague, S.J.; Angew. Chem. Int. Ed. 1999, 38, 736 - 749
P3
Tyr

core(1)-R2 [Xe][C@@]([H])(C)C#N
[Xe][C@]([H])(C#N)
C(C)(C)C
[Xe][C@]([H])(C)C#
N [Xe]C1(CC1)C#N
[Xe]C1(CCN(CC1)C)C
#N [Xe]CC#N
[Xe]N1CCC(CC1)Cc2ccccc2
7.8 0.0 6.4 8.3 7.6 8.1
[Xe]N1CCC(CC1)Oc2ccc(cc2)F
6.9 4.0 4.8 4.9 0.0 7.3
[Xe]N1CCc2c(c3cc(ccc3[nH]2)F)C1
7.8 6.1 7.8 9.8 0.0 9.2
[Xe]N1CCc2c(ccs2)C1
8.5 0.0 6.9 8.6 7.7 8.9
[Xe]N1CCc2ccc(cc2C1)F
8.7 5.5 6.8 8.5 0.0 8.7
[Xe]N1CCc2ccccc2C1
8.5 0.0 7.2 8.8 7.6 8.8
[Xe]N1CCCCC1
0.0 0.0 5.2 6.7 0.0 7.1
[Xe]N1CCN(CC1)c2ccc(cc2)F
0.0 5.4 6.2 7.9 0.0 7.6
[Xe]N1CCN(CC1)c2ccc(cc2)S(=O)(
=O)C
7.1 4.9 6.4 6.7 4.8 8.1
[Xe]N1CCN(CC1)c2nc3ccccc3s2
7.8 0.0 6.3 8.2 6.8 8.0
Xe
N
Xe
N
O
F
Xe
N
N
F
Xe N
S
Xe
N
F
Xe
N
Xe N
Xe N N F
Xe N N S
O
O
Xe N N
N
S
Xe
H
N
Xe H
N
Xe
H
N
Xe
N
Xe
N
N Xe N
7.7
7.3
Free-Wilson – P1 and P3 groups combinations
4.04.0
5.1
5.1
5.4
5.3
NO NO
9.8
8.2
• Whole team made effort
• 3 weeks
• 54 out of 60 made
• SAR, SPR, in-vitro DMPK
• Cat K, B, L and S selectivity
• Rat PK on 10 compounds
• Two series had best DMPK
and selectivity profiles
• In-silicon models improved
• AZLogD good
• hERG and Solubility

F
N
N
O
N O
N
O
O
N
NO
N
S
N
O
O
ON
N
O
N
N
F
ON
N
O
N
NO
N
S
O
N
NO
N
N
Piperazines Carbolines
Cat K pIC50 6.5
LogD7.4 <-0.5
LLE >6.0
Med Chem route to AZD4996
AZ11961622 Hit
Cat K pIC50 8.7
AZ12475452 Lead
Cat K pIC50 7.95
DTM ~1.0 mg/kg UID
All Renal Clearance
AZ12581322
Cat K pIC50 8.0
DTM 0.5 mg/kg/UID
AZ12578219
Cat K pIC50 9.0
DTM 1.0 mg/kg/BID
AZ12657125
Cat K pIC50 9.1
DTM 0.05 mg/kg/UID
Dossetter, A.G. et al Bioorg Med Chem Lett. 2012, 22(17), 5563-5568.
Dossetter, A.G. et al J Med Chem. 2012, 55(14), 6363-6374.
Unusual structural change

Piperazine serie CD2/3
pIC50 7.95
LogD 0.67
HLM <2.0
Solubility 280µM
DTM ~1.0 mg/kg UID
Potent
Too polar / Renal Cl
PDB - 97% of structures
Crawford, J.J.; Dossetter, A.G J Med Chem. 2012, 55, 8827.
Dossetter, A. G. Bioorg. Med. Chem. 2010, 4405
pIC50 8.2
LogD 2.8
HLM <1.0
Solubility >1400µM
DTM 0.01 mg/kg UID
High F% / stability
maximised
Increase in LogP,
Properties improved
Solubility
DpIC50 - 0.1
DLogD +1.4
DpSol +1.2
DHLM + 0.25
No renal Cl
low F%
DpIC50 +0.1
DLogD - 0.7
DpSol ~0.0
DHLM - 0.25
High F%
rat/DogElectrostatic potential minima between oxygens
Approx like N from 5-het, new compound can not
form a quinoline
Incr. selectivity
DpIC50 +0.1
DLogD - 0.7
DpSol ~0.0
DHLM - 0.25
High F%
rat/Dog

Not for Circulation
Project example
11-bHSD
Understanding the data you have
CNS penetration properties
3D – Structure based design and TPSA usage
Goldeberg, Dossetter et al J. Med. Chem. 2014,
57, 970−986 dx.doi.org/10.1021/jm4016729

State of play 1Q 2010:
AZ CDs moving through pre-clinical studies – only weak CNS penetration
Aiming for “diabetes+” product profile
!Work started Feb 2010 to design a better CNS penetrant compound
– goal higher glycaemic control + bodyweight loss/maintenance
11-bHSD Project Background
N
O
N
N
HO2C
NO
N
N
N
O
OH N
N
N
N
O
O
O
OH
CD2 backup
peripherally restricted
short t1/2
CD2
peripherally restricted
long t1/2
CNS probe
weakly CNS penetrant
short t1/2

%WeightChange
In-vivo evidence
AZ13078072 / AZ1638301 in 48% HFF DIO: PO
Good evidence that CNS exposure is required for Glucose & BW activity
Efficacy in BW was achieved in DIO mice with Brain free IC50 cover >10x
Evidence for T1/2 of ~10 hours in human required
Would improved CNS penetration increase efficacy?
7.5 mg/kg BID 20 mg/kg BID
%WeightChange
AZ12638301

“Lipinski Rule of Four”
Further analysis of World Drug Index of CNS drug reveals that for
penetration through the Blood Brain Barriers:-
Mol. Weight <400
ClogP <4
HB-Donors <4
HB-Acceptors <8
More recent studies and understanding
TPSA <125
MDCK Efflux None (surrogate for active transport)
Di, L.; Rong, H.; Feng, B. Demystifying brain penetration in central nervous system drug discovery.
J. Med. Chem. 2013, 56, 2−12.
Desai, P. V.; Raub, T. J.; Blanco, M.-J. How hydrogen bonds
impact P-glycoprotein transport and permeability. Bioorg. Med. Chem. Lett. 2012, 22, 6540−6548.
Heffron, T. P. et al. The design and identification of brain penetrant inhibitors of phosphoinositide
3- kinase α. J. Med. Chem. 2012, 55, 8007−8020.
Factors that Influence CNS Absorption

Scaffold hops – new cores optimised for CNS
1) Use old data to
identify optimal
known core for LE,
make best cpds and
profile whilst exploring
removal of H bond
donors on warhead
2) Make ~20 new
core systems with
goal of identifying
best cores and
profile for CNS PK
N OH
O
N
N
Pyrazoles
OHN
R2
O
S
N
R1
OHN
O
N
N
N
R1
R2
R3
Thiazoles
PPs
3) Optimise best cores to
identify CD and bioscience
probe to prove CNS
hypothesis
OHN
O
A
A
A
A
A
R1
"
R2
"

Learning: role of the pyridyl nitrogen and ethers
!No obvious interaction with protein
!Hypothesis: nitrogen locks
compound into bioactive
conformation
!pIC50/LLE disconnect due to
competing desolvation effects?
! Stopped PPs ‘3166 best comp
N OH/CN
O
N
N
N
C5
C6
C7
Wide Scope
Grease
Polarity
(NADPH)
Grease
Key
Interaction
Ab initio acceptor strength correlates
to LLE (but not pIC50)
N gives you +0.7 log units
potency (logD neutral)
Graeme Robb
N
O
N
N
N
O
CN
Hu=7.9
Mu=7.9
logkB on ethers is
conformation dependent!
CN
OH
AZ13503166
pIC50 7.9 hum
pIC50 7.9 Mo
Free Br / Bl 1.01
Sol 7.8 uM
Dog Heps 28

CNS – Descriptors- How do they get in?
• TPSA best indicator
• Maybe size is
important?
95 A2 PSA
Ertl, Rohde, Selzer “Fast Calculation of Molecular Polar Surface Area as a Sum of Fragment-
Based Contributions and Its Application to the Prediction of Drug Transport Properties”:
J.Med.Chem. 2000, 43, 3714-3717

Log D7.4 and in-vitro / in-vivo clearance
Author | 00 Month Year43 Set area descriptor | Sub level 1
No relationship

Optimising thiazoles for mouse – targeting Asp259
O
NAdOH
S
N
MeO
O
NAdOH
S
N
MeO
O
O
NAdOH
S
N
O
MeO O
NAdOH
S
N
O
O
!Thiazoles can give no drop-off between Hu
and Mu with extended ethers
!Asp259 (water-mediated) believed to be
important
!Bigger cpds - do we want to do this for
CNS?
! Expanded set with MPS
Asp259

Conformation of C4 groups
• Cyclobutyl
– minE is reverse
conf
– Bound is
+1.1kcal/mol
• Cyclopropyl
– minE is reverse
conf
– Bound is
+1.3kcal/mol
• THF
– minE is bound conf
– -2.8kcal/mol from
reverse
• Torsional strain plots – where is the energy minimum?
Orange line is
accurate calc
Favourable combination of thiazole and C2 THF group

Free-Wilson BHSD Thiazole Fabs
• Splits Fabs categories based on R-
groups
• 1st split on R2:
- Bad:
• 2nd split on R4:
- Bad:
• Therefore good groups are:
0.00
0.25
0.50
0.75
1.00
Fabscat
All Rows
Run 1 : R1 SMILES([R1][C
...
Run 1 : R1 SMILES([R1]OC1CCOCC1, [R1]OC
...
Run 1 : R2 SMIL
...
Run 1 : R2 SMILES([R2]C1CC
...
1. >0.33
2. <0.33
Included
Excluded
0.835
-2.18
RSquare
22
3
N
2
Number
of Splits
All Rows
22
Count
30.316406
G^2
1.2531897
LogWorth
Run 1 : R1
SMILES([R1][C@H]1CCOC1, [R1]C,
[R1]O[C@@H]1CCOC1, [R1]OCCO)
8
Count
0
G^2
Run 1 : R1
SMILES([R1]OC1CCOCC1, [R1]OC,
[R1]COC, [R1]C1CCOC1,
[R1]OC(C)C, [R1]OC1COC1,
[R1]OCC)
14
Count
16.751548
G^2
1.6062913
LogWorth
Run 1 : R2
SMILES([R2][C@@H]1CCCO1,
[R2][C@]1(CCCO1)[H],
[R2]CC(F)(F)F, [R2][C@H]1CCCO1)
5
Count
5.0040242
G^2
Run 1 : R2 SMILES([R2]C1CC1,
[R2]C1CCC1, [R2]OC)
9
Count
0
G^2
Partition for Fabs cat
AdamOH

Combining the groups gave a long shortlist…..
Compound
Name
AZ13556560 AZ13546532 AZ13507701 AZ13546293
Hum E/10 HPLC data / uM 0.0004 0.0002 0.0004 0.0004
Human PPB (% free) 9.5 21 18 21
CNS? Total Brain / Blood
Ratio
0.31 0.34 1.07 0.9
Free Brain/Blood
Ratio
0.23 0.14 0.74 0.51
Rat Blood
F (%)
23-68 100 5-40 15
Compound
Name
AZ13546285 AZ13554465 AZ13574806 AZ13540510
Hum E/10 HPLC data / uM 0.0005 0.0005 0.0007 0.0017
Human PPB (% free) 39 17 28 23
CNS? Total Brain / Blood
Ratio
0.83 1.12 0.94 1.14
Free Brain/Blood
Ratio
0.98 1.04 0.88
Rat Blood
F (%)
24 (0.4 Fabs) 33 (0.7 Fabs) 87 (1.0 Fabs) 29-83
N
O
O
O
N
N
O
O O
N O
O
S
N
N O
O
S
N
O
O
H
O
N O
O
S
N
O
O
HO
N O
O
S
N
O
O
ON
O
S
N
ON
N O
O
S
N
O
O
O
H
N O
S
N
O
O
O
H

Not for Circulation
Conclusions
What to do, what not to do….

Not for circulation
Key conclusions
• We use In-Silico techniques to refine our compound designs and achieve our goal of a quality
candidate drug in fewer iterations.
• Get as much information as you can..
– SAR from literature / patents / structures
• good informatics tools are available now.
• Free Wilson (Permutative MMPA) can be highly nformative
• Quality analysis of what you have already have is key
– 2D plots, same scale x/y axis, measured data on a continuous scale
• 2D Design - Prediction methods
– Understand how “models” are built what their limitations are
• Descriptions, atom or group patterns are used to encode molecules (AlogP98)
• Model are only as good as the dataset from which it was generated
• Every prediction has an error – understanding how big this effects the decision
– Newer AI methods can be used to generate new idea molecules
• 3D Design / Structure Based Drug Design (SBDD)
– Rapid 3D conformation generation is based on look up tables and fast minimization –
approximate only
– Docking methods often use 3D conforms (or poses) – both have error – errors add up
– QM techniues can given superior 3D shape but take time to compute
– Remember a small molecule xtal structure of a molecule can be very informative
– PDB and CCDB is a treasure-trove of information
• don’t re-invent the wheel – avoid having to calculate
– Remember a crystal structure of a small molecule is a binding pocket is just a snapshot
– It is a model built from the electron density found (if RMSE is high and density is poor don’t use it)

Not for Circulation
About MedChemica
>10 experience in building A.I. Systems for drug discovery

Not for circulation
• Founded in 2012 by experienced large Pharma
medicinal/computational chemists to accelerate drug
hunting by exploiting data driven knowledge
• Domain leaders in SAR knowledge extraction and
knowledge based design
• > 10 years experience of building AI systems that suggest
actions to chemists (6 years as MedChemica)
• Creators of largest ever documented database of
medicinal chemistry ADMET knowledge
MedChemica Publications

Not for circulation
…7 Years of working with pharma
companies
“Our median number of compounds per LO project is 3000 - this is
unsustainable… [it should be] 300”
– Director of Chemistry (large pharma)
“Can we define the text book of medincal chemistry?”
– Director of Comp Chem (large pharma)
“We are aiming at 300 compound per project – currently we are
about 400, we will get better”
– ExScienta scientist at SCI ‘What can BigData do for chemistry’ –
London Oct 2017
MedChemica is a company using knowledge extraction techniques
to build “expert systems” to suggest actions to chemists [Artificial
Intelligence – AI] and reduce the time and cost to critical compounds
and candidate drugs.

Not for circulation
AI Software Platforms
– Complete In-house platform
– Analysis of own data and automated
updating
– Design tool access to all chemists
– Custom fitting (Software-as-a-Service)
One stop GUI
Design tool
Biotech, Universities and
Foundations
Medium to large pharma,
agrochemical and
materials research
– Secure web-based AI design platform
– CHEMBL, Patent data analysed
– Merged into one knowledgebase

Browser / Command line In House tools Well known design tools
Pair, Rule, Model
Database
RuleDesign
Exploitation Capabilities
RESTful API
SAR PairsSpotDesign™
Matched Molecular Pair Analysis and
Machine Learning
Corporate DB
AI ready
Structures and
Data
MCPairs
Server

Not for circulation
Science As A Service (SaaS)
Target ID
Hit
Screening
Lead Identification Lead Optimisation Pre-Clinical
AI H2L design
sets
Bespoke Advanced Analytics and Computational Chemistry services through-out the research phase
Compound design to
solve ADMET and
potency issues
Third party
compound
assessment
Directed virtual screening
for hit matter
Library design for novel
protein targets
AI Toxophore
assessment
Patent analysis
Pharmacophore
profiling
Generating IP for
clients
[Scaffold hops]
Collection
evaluation
and
enhancement
CLICK
FOR
CASE
STUDIES

Not for circulation
Benefits of MCPairs Enterprise and Online
• Data driven compound design from extensive knowledge database
• Fewer compounds to make, faster to project results
• Proven results from more than 21 organisations / projects (1 failure)
• Collective 50 years of drug discovery experience on top of AI
• Backed by 10 experience of building “AI systems”
• Advanced analysis of patent data and potency prediction
• Novel compound generation and improvement

Not for circulation
Experience with Enterprise Clients
6060

MDC Connect: In-Silico Drug Design - what to do, what not to do - project driven examples

More Related Content

What's hot (20)

Similar to MDC Connect: In-Silico Drug Design - what to do, what not to do - project driven examples (20)

More from Medicines Discovery Catapult (20)

Recently uploaded (20)

MDC Connect: In-Silico Drug Design - what to do, what not to do - project driven examples