SlideShare a Scribd company logo
MedChemica
MedChemica
For Example
Does adding a 4-F to a phenyl ring actually reduce
metabolism?
Mean change in human liver microsome log(Clint).
497 examples Mean (m) = 0.06, SE = 0.02
Probability change >0.5 log improvement : 0.086 = 8.6%
NO – the evidence is that 4-H>>4-F on average will give no difference
and only 8.6% of the time will it make a significant difference.
Medicinal Chemistry
transformations
Classic changes on med chem?
How often do they work?
How can we test the reliability of the
“rules” of medicinal chemistry?
Multiparameter Optimization of Pharmaceuticals: the Big-Data Way
Alexander G. Dossetter, Andrew G. Leach, Edward J. Griffen.
al.dossetter@medchemica.com
Ebenezer House, Ryecroft, Newcastle-under-Lyme, Staffordshire, ST5 2BE
Can we analyse multi-pharma datasets to produce the largest electronic textbook of Medicinal Chemistry?
If so can we reduce the number of compounds we make (costs) and increase quality?
Does it work? Multi-Objective Optimisation using Knowledge base - Project Examples
Identify Matched Pairs
in a supervised manner
on a dataset
Results Log10
and
subtracted
OR via an algorithm
that finds all pairs in
large datasets
Matched Molecular Pair Analysis
(MMPA)
Diversity testing
n = number of matched pairs
Statistical analysis (μ,σ,SEM)
frequency of increase
frequency of increase > 0.5
Main focus: In vitro ADMET assays
KNOWLEDGE BASE
Coded transforms
+ plus data
analysis
- Novel more efficient core required, improve hERG for CD
- CNS penetration and deliver tool for in vivo testing
McCoull, Dossetter et al, Med. Chem. Commun., (2013), 4, 456
Ghrelin Inverse agonists
MMPA
Cores
- Fix hERG problem whilst maintaining potency
Waring et al, Med. Chem. Commun., (2011), 2, 775
Glucokinase Activators
MMPA
∆pEC50: -0.1 ∆logD: -0.6 ∆hERG pIC50 :-0.5
n=33 n=32 n=22
MMPA
∆pEC50: +0.3 ∆logD: +0.3 ∆hERG pIC50 :-0.3
n=20 n=23 n=19
MMPA
∆pEC50: -0.1 ∆logD: -0.6 ∆hERG pIC50 :-0.5
n=27 n=27 n=7
Gleeson, 2009 and Papadatos, 2010
Human Liver Microsomal stabilityGroups that reduce cytochrome P450
inhibition or hERG ion channel binding
A1 = alkyl
A1 = aryl
A1 = all
What sort of Design Rules do you get? Transforms from the MMPA literature - Dossetter, A.G.; Griffen, E.J.; Leach A.G. Drug Discov. Today 2013, 18, 724
Dossetter 2012
Human Liver Microsomal stability
R1 = N, O, S, alkyl, arom. R2 = alkyl, arom.
Dossetter 2012
We can make it be better: More data will mean more specific transforms, with the best chance of delivering progress
• Distribution of clogP change for
18484 transformations where
hERG pIC50 decreases by at
least 0.3 log units and n >=3
• 25% of the hERG transforms
above show clogP >0
- Rules from analysis of 60K hERG datapoints
- Matched Pair Enantiomers produce different results
Surprises from MMPA – its more than lipophilicity
hERG If physical properties drove ADMET
then enantiomeric pairs should be
equivalent
Enantiomeric pairs reveal that key medicinal
chemistry parameters differ between R and S

Andrew G. Leach et al,
Med. Chem. Commun., 2012,3, 528-540.
6 Great Things to do next
1. Make better molecules faster – more data better rules
2. Generate new rules by suggesting extra compounds to test – fill in
the gaps
3. Critical toxicology SARs – more assays analyzed
4. Understand species differences in clearance
5. Combine rules to create “meta-rules”
6. Combine with shape and color to generate pharmacophores
7800k
Unique Transforms
1200k n>1
218k n>4
73k n>9
Pfizer authors extracted their HLM
dataset of 225k results – only a
fraction (<1%) had enough MPs to
reach statistical significance
Keefer, C.E. et al. Bioorg. Med. Chem. (2011)
19, 3739–3749
Bayer and AZ compound collections
overlap by just 5%!
Kogej, T. et al. Drug Discov. Today. 2012
There is so much we don’t know about
Combining data will find more rules
-1.7-1.4-1.1-0.8-0.5-0.20 .2 .4 .6 .8 1 1.2 1.5
pIC50 -0.4
logD -1.8
hERG pIC50 +0.4
pIC50 9.9
logD 5.0
hERG pIC50 5.0
LLE 4.9
very potent
very lipophilic
pIC50 +0.9
logD +0.2
hERG pIC50 -0.3
pIC50 8.2
logD 1.3
hERG pIC50 4.4
LLE 6.9
pIC50 -2.2
logD -2.2
hERG pIC50 -0.7
100
compounds
made
LLE 6.4
LLE 6.9
Better Worse Indistinguishable
Conclusions
• Projects can be accelerated by delivering statistically sound medicinal chemistry knowledge.
• The design rules found can be counter-intuitive to current ideas in medicinal chemistry.
• Pooling data can provide more rules including more structurally specific ones.
• By focusing on ADMET, rules are found that help reduce efficacy and potency failures in the clinic.

More Related Content

Viewers also liked

Presentation media
Presentation media Presentation media
Presentation media
University of Winchester
 
Principios de Diseño de Realidad Virtual
Principios de Diseño de Realidad VirtualPrincipios de Diseño de Realidad Virtual
Principios de Diseño de Realidad Virtual
UX Nights
 
Manual de usuario_de_gimp_
Manual de usuario_de_gimp_Manual de usuario_de_gimp_
Manual de usuario_de_gimp_
Informática Liceo
 
Exmarko
ExmarkoExmarko
Exmarko
stjosnelson
 
LifeHackDay 2016 - Odessa: Дима Гадомский, Axon Partners
LifeHackDay 2016 - Odessa: Дима Гадомский, Axon PartnersLifeHackDay 2016 - Odessa: Дима Гадомский, Axon Partners
LifeHackDay 2016 - Odessa: Дима Гадомский, Axon Partners
Sergey Dovgopolyy
 
Scottish Independence Referendum: 200 days to go
Scottish Independence Referendum: 200 days to goScottish Independence Referendum: 200 days to go
Scottish Independence Referendum: 200 days to go
Ipsos UK
 
Reflections on Dhammapada Verses
Reflections on Dhammapada VersesReflections on Dhammapada Verses
Reflections on Dhammapada Verses
OH TEIK BIN
 
Empreendedorismo
EmpreendedorismoEmpreendedorismo
Empreendedorismo
Wandick Rocha de Aquino
 
RWDG Webinar: A Data Governance Framework for Smart Data
RWDG Webinar: A Data Governance Framework for Smart DataRWDG Webinar: A Data Governance Framework for Smart Data
RWDG Webinar: A Data Governance Framework for Smart Data
DATAVERSITY
 
Enfermedades neurológicas crónicas en españa.
Enfermedades neurológicas crónicas en españa.Enfermedades neurológicas crónicas en españa.
Enfermedades neurológicas crónicas en españa.
José María
 
BOOST YOUR CONFIDENCE
BOOST YOUR CONFIDENCEBOOST YOUR CONFIDENCE
BOOST YOUR CONFIDENCE
Ivan Williams
 
20 min fusion day persona talk
20 min fusion day persona talk20 min fusion day persona talk
20 min fusion day persona talk
Ramya Mahalingam
 

Viewers also liked (12)

Presentation media
Presentation media Presentation media
Presentation media
 
Principios de Diseño de Realidad Virtual
Principios de Diseño de Realidad VirtualPrincipios de Diseño de Realidad Virtual
Principios de Diseño de Realidad Virtual
 
Manual de usuario_de_gimp_
Manual de usuario_de_gimp_Manual de usuario_de_gimp_
Manual de usuario_de_gimp_
 
Exmarko
ExmarkoExmarko
Exmarko
 
LifeHackDay 2016 - Odessa: Дима Гадомский, Axon Partners
LifeHackDay 2016 - Odessa: Дима Гадомский, Axon PartnersLifeHackDay 2016 - Odessa: Дима Гадомский, Axon Partners
LifeHackDay 2016 - Odessa: Дима Гадомский, Axon Partners
 
Scottish Independence Referendum: 200 days to go
Scottish Independence Referendum: 200 days to goScottish Independence Referendum: 200 days to go
Scottish Independence Referendum: 200 days to go
 
Reflections on Dhammapada Verses
Reflections on Dhammapada VersesReflections on Dhammapada Verses
Reflections on Dhammapada Verses
 
Empreendedorismo
EmpreendedorismoEmpreendedorismo
Empreendedorismo
 
RWDG Webinar: A Data Governance Framework for Smart Data
RWDG Webinar: A Data Governance Framework for Smart DataRWDG Webinar: A Data Governance Framework for Smart Data
RWDG Webinar: A Data Governance Framework for Smart Data
 
Enfermedades neurológicas crónicas en españa.
Enfermedades neurológicas crónicas en españa.Enfermedades neurológicas crónicas en españa.
Enfermedades neurológicas crónicas en españa.
 
BOOST YOUR CONFIDENCE
BOOST YOUR CONFIDENCEBOOST YOUR CONFIDENCE
BOOST YOUR CONFIDENCE
 
20 min fusion day persona talk
20 min fusion day persona talk20 min fusion day persona talk
20 min fusion day persona talk
 

More from Al Dossetter

How we Built a Large Scale Matched Pair Analysis Engine (MCPairs) using OpenE...
How we Built a Large Scale Matched Pair Analysis Engine (MCPairs) using OpenE...How we Built a Large Scale Matched Pair Analysis Engine (MCPairs) using OpenE...
How we Built a Large Scale Matched Pair Analysis Engine (MCPairs) using OpenE...
Al Dossetter
 
Practical Drug Discovery using Explainable Artificial Intelligence
Practical Drug Discovery using Explainable Artificial IntelligencePractical Drug Discovery using Explainable Artificial Intelligence
Practical Drug Discovery using Explainable Artificial Intelligence
Al Dossetter
 
MedChemica Active Learning - Combining MMPA and ML
MedChemica Active Learning - Combining MMPA and MLMedChemica Active Learning - Combining MMPA and ML
MedChemica Active Learning - Combining MMPA and ML
Al Dossetter
 
Accelerating multiple medicinal chemistry projects using Artificial Intellige...
Accelerating multiple medicinal chemistry projects using Artificial Intellige...Accelerating multiple medicinal chemistry projects using Artificial Intellige...
Accelerating multiple medicinal chemistry projects using Artificial Intellige...
Al Dossetter
 
MedChemica - Automated Extraction of Actionable Knowledge from Large Scale in...
MedChemica - Automated Extraction of Actionable Knowledge from Large Scale in...MedChemica - Automated Extraction of Actionable Knowledge from Large Scale in...
MedChemica - Automated Extraction of Actionable Knowledge from Large Scale in...
Al Dossetter
 
MedChemica BigData What Is That All About?
MedChemica BigData What Is That All About?MedChemica BigData What Is That All About?
MedChemica BigData What Is That All About?
Al Dossetter
 
MedChemica - Extracting and exploiting medicinal chemistry ADMET knowledge au...
MedChemica - Extracting and exploiting medicinal chemistry ADMET knowledge au...MedChemica - Extracting and exploiting medicinal chemistry ADMET knowledge au...
MedChemica - Extracting and exploiting medicinal chemistry ADMET knowledge au...
Al Dossetter
 

More from Al Dossetter (7)

How we Built a Large Scale Matched Pair Analysis Engine (MCPairs) using OpenE...
How we Built a Large Scale Matched Pair Analysis Engine (MCPairs) using OpenE...How we Built a Large Scale Matched Pair Analysis Engine (MCPairs) using OpenE...
How we Built a Large Scale Matched Pair Analysis Engine (MCPairs) using OpenE...
 
Practical Drug Discovery using Explainable Artificial Intelligence
Practical Drug Discovery using Explainable Artificial IntelligencePractical Drug Discovery using Explainable Artificial Intelligence
Practical Drug Discovery using Explainable Artificial Intelligence
 
MedChemica Active Learning - Combining MMPA and ML
MedChemica Active Learning - Combining MMPA and MLMedChemica Active Learning - Combining MMPA and ML
MedChemica Active Learning - Combining MMPA and ML
 
Accelerating multiple medicinal chemistry projects using Artificial Intellige...
Accelerating multiple medicinal chemistry projects using Artificial Intellige...Accelerating multiple medicinal chemistry projects using Artificial Intellige...
Accelerating multiple medicinal chemistry projects using Artificial Intellige...
 
MedChemica - Automated Extraction of Actionable Knowledge from Large Scale in...
MedChemica - Automated Extraction of Actionable Knowledge from Large Scale in...MedChemica - Automated Extraction of Actionable Knowledge from Large Scale in...
MedChemica - Automated Extraction of Actionable Knowledge from Large Scale in...
 
MedChemica BigData What Is That All About?
MedChemica BigData What Is That All About?MedChemica BigData What Is That All About?
MedChemica BigData What Is That All About?
 
MedChemica - Extracting and exploiting medicinal chemistry ADMET knowledge au...
MedChemica - Extracting and exploiting medicinal chemistry ADMET knowledge au...MedChemica - Extracting and exploiting medicinal chemistry ADMET knowledge au...
MedChemica - Extracting and exploiting medicinal chemistry ADMET knowledge au...
 

Multi-Parameter Optimization of Pharmaceuticals: the Big-Data Way - Poster 36 Cambridge Med Chem

  • 1. MedChemica MedChemica For Example Does adding a 4-F to a phenyl ring actually reduce metabolism? Mean change in human liver microsome log(Clint). 497 examples Mean (m) = 0.06, SE = 0.02 Probability change >0.5 log improvement : 0.086 = 8.6% NO – the evidence is that 4-H>>4-F on average will give no difference and only 8.6% of the time will it make a significant difference. Medicinal Chemistry transformations Classic changes on med chem? How often do they work? How can we test the reliability of the “rules” of medicinal chemistry? Multiparameter Optimization of Pharmaceuticals: the Big-Data Way Alexander G. Dossetter, Andrew G. Leach, Edward J. Griffen. al.dossetter@medchemica.com Ebenezer House, Ryecroft, Newcastle-under-Lyme, Staffordshire, ST5 2BE Can we analyse multi-pharma datasets to produce the largest electronic textbook of Medicinal Chemistry? If so can we reduce the number of compounds we make (costs) and increase quality? Does it work? Multi-Objective Optimisation using Knowledge base - Project Examples Identify Matched Pairs in a supervised manner on a dataset Results Log10 and subtracted OR via an algorithm that finds all pairs in large datasets Matched Molecular Pair Analysis (MMPA) Diversity testing n = number of matched pairs Statistical analysis (μ,σ,SEM) frequency of increase frequency of increase > 0.5 Main focus: In vitro ADMET assays KNOWLEDGE BASE Coded transforms + plus data analysis - Novel more efficient core required, improve hERG for CD - CNS penetration and deliver tool for in vivo testing McCoull, Dossetter et al, Med. Chem. Commun., (2013), 4, 456 Ghrelin Inverse agonists MMPA Cores - Fix hERG problem whilst maintaining potency Waring et al, Med. Chem. Commun., (2011), 2, 775 Glucokinase Activators MMPA ∆pEC50: -0.1 ∆logD: -0.6 ∆hERG pIC50 :-0.5 n=33 n=32 n=22 MMPA ∆pEC50: +0.3 ∆logD: +0.3 ∆hERG pIC50 :-0.3 n=20 n=23 n=19 MMPA ∆pEC50: -0.1 ∆logD: -0.6 ∆hERG pIC50 :-0.5 n=27 n=27 n=7 Gleeson, 2009 and Papadatos, 2010 Human Liver Microsomal stabilityGroups that reduce cytochrome P450 inhibition or hERG ion channel binding A1 = alkyl A1 = aryl A1 = all What sort of Design Rules do you get? Transforms from the MMPA literature - Dossetter, A.G.; Griffen, E.J.; Leach A.G. Drug Discov. Today 2013, 18, 724 Dossetter 2012 Human Liver Microsomal stability R1 = N, O, S, alkyl, arom. R2 = alkyl, arom. Dossetter 2012 We can make it be better: More data will mean more specific transforms, with the best chance of delivering progress • Distribution of clogP change for 18484 transformations where hERG pIC50 decreases by at least 0.3 log units and n >=3 • 25% of the hERG transforms above show clogP >0 - Rules from analysis of 60K hERG datapoints - Matched Pair Enantiomers produce different results Surprises from MMPA – its more than lipophilicity hERG If physical properties drove ADMET then enantiomeric pairs should be equivalent Enantiomeric pairs reveal that key medicinal chemistry parameters differ between R and S 
Andrew G. Leach et al, Med. Chem. Commun., 2012,3, 528-540. 6 Great Things to do next 1. Make better molecules faster – more data better rules 2. Generate new rules by suggesting extra compounds to test – fill in the gaps 3. Critical toxicology SARs – more assays analyzed 4. Understand species differences in clearance 5. Combine rules to create “meta-rules” 6. Combine with shape and color to generate pharmacophores 7800k Unique Transforms 1200k n>1 218k n>4 73k n>9 Pfizer authors extracted their HLM dataset of 225k results – only a fraction (<1%) had enough MPs to reach statistical significance Keefer, C.E. et al. Bioorg. Med. Chem. (2011) 19, 3739–3749 Bayer and AZ compound collections overlap by just 5%! Kogej, T. et al. Drug Discov. Today. 2012 There is so much we don’t know about Combining data will find more rules -1.7-1.4-1.1-0.8-0.5-0.20 .2 .4 .6 .8 1 1.2 1.5 pIC50 -0.4 logD -1.8 hERG pIC50 +0.4 pIC50 9.9 logD 5.0 hERG pIC50 5.0 LLE 4.9 very potent very lipophilic pIC50 +0.9 logD +0.2 hERG pIC50 -0.3 pIC50 8.2 logD 1.3 hERG pIC50 4.4 LLE 6.9 pIC50 -2.2 logD -2.2 hERG pIC50 -0.7 100 compounds made LLE 6.4 LLE 6.9 Better Worse Indistinguishable Conclusions • Projects can be accelerated by delivering statistically sound medicinal chemistry knowledge. • The design rules found can be counter-intuitive to current ideas in medicinal chemistry. • Pooling data can provide more rules including more structurally specific ones. • By focusing on ADMET, rules are found that help reduce efficacy and potency failures in the clinic.