Call Girls Haridwar Just Call 8250077686 Top Class Call Girl Service Available
The End of the Drug Development Casino?
1. Machine Learning: The End of the Drug
Development Casino?
Paul Agapow
Data-Driven Discovery and Development in Pharma R&D
Public
October 2020
2. Disclosure
• Based on experience in current &
previous positions
• Oncology R&D RWE / ML&AI @AZ
• Data Science Institute @ICL
• Centre for Infection @HPA(UK)
• Does not reflect official AZ thought or
projects
• No conflicts of interest
2
4. There is a revolution in drug development …
• Every day we hear of new
advances & developments
• Acceleration in basic
biomedical research
• Constant development of
new molecular
technologies
• An age of cheap
computation & powerful
machine learning
4
5. • It costs ~ $1-2B and 10 years to
develop & launch a drug
• Each patient in a clinical trial costs
$1-10K
• The “valley of death”: most
candidate drugs will fail
• Post-approval adds to the costs
5
… Yet the maths of drug development remain tough
ePharmacology.hubpages.com
6. … And costs continue to increase
• Eroom’s Law: cost of
developing new drug roughly
doubles every nine years
• Acceleration of biomedical
research not reflected in drug
development
• Recent uptick in approvals
does not reflect decreasing
costs
6
Pharmacelera (2014)
7. What is driving the downwards trend?
• Unclear but:
• All the low-hanging fruit has been
picked
• Bar being raised – every drug has
to beat the previous one
• Increasing targeting of smaller
subpopulations
• Increasingly exotic nature of
drugs
• …
7
Forbes (2017)
8. Drug development is a numbers game.
And the numbers are (increasingly) not good.
How do we tilt the odds in our favour?
Are Data Science, Machine Learning and AI the
answer?
Where & how?
8
10. As data science, ML, AI (etc.) are such
murky terms:
• Low assumption, non-parametric
approaches
• Low on explicit models
• Derived from data and improved by
more data
• On a spectrum with statistical inference
Well-suited to drug dev problems, a force
multiplier
10
What do I mean by Machine Learning? Or Data Science?
Google (2018)
11. • Landmark AZ papers:
• Cook et al. 2014
• Morgan et al. 2018
• 5 Rs:
• right target
• right patient
• right tissue
• right safety
• right commercial potential
11
We neglect the critical (and obvious) problems of drug development
12. To be effective, we have to solve the right problems …
“Academic”
• Interesting problems
• Low-hanging fruit
• Ideal, clean data
• Problems we have data for
• Isolated, simple biology
• “Proof of concept”
• Focus on early dev
12
“Industrial”
• Needful problems
• Constant need to raise bar
• Real, messy data
• Often ill-defined problems
• Real, systemic biology
• Operational
• Need help in late dev
13. • RCTs account for a large part of the
cost of drug development
• More trials == more money
• Up to 85% of trials fail to find
enough patients
• Underpowered trials are harder to
interpret
• Trials costs largely down to
inflexible expenses (labs, admin)
Caveat: as they currently stand
13
Running more clinical trials as they currently exist is not a solution
ClinicalResearch IO (2018)
15. • A tendency to treat drug
development as just a data problem
• Tendency towards simplified, isolated
biology
• Due to ML & shift to high
throughput methods
• But disease is systemic
• And patients are complex
• And the necessary data is
complex
15
Recognise that biology isn’t “just the domain”
16. We need to work outside ‘table-land’
• A patient / disease is more than a
table, data can be:
• Connected
• Multi-modal
• Sparse and/or incomplete
• Examples:
• Integrative analysis / multi-omics
• Graph methods incl. graph
neural networks
16
We need methodology for complex biology & complex data
Clinical asthma types are reshaped & refined
by transcriptomic and proteomic data layers
Kermani et al. (2020)
17. Examples: disease and patient graphs mined from EHRs
17
COPD T2D
Transform patients into sequences of diagnosis
codes
Look for over-represented temporal pairs of codes
Collapse pairs into trajectories of diagnoses
Combine similar trajectories with graph similarity
Brunak et al. Nature Coms. 2016
Topology based Patient-Patient network, identify
distinct subtypes of T2D
Dudley et al. Sci. transl. Med, 2015
18. • Deep learning / neural networks have
shown amazing success
• But conventional architectures ill-suited
to many biomedical problems:
• Fixed width inputs
• Neglect relationships
• Difficulties with multiple data
modalities
• Graph convolutional networks &
knowledge graphs
• Model relationships
• Integrate different data sources
18
Example: Graph-based neural networks
DeePaN: stratifying NSCLC patients likely to
benefit from IO therapy
Fang et al. (2017)
19. • Similar patient presentation can
mask vastly different molecular
machinery
• Even within a “homogenous”
condition, patients will have
different outcomes
• What are the treatment effects for
individual patients?
Understanding these leads to:
• More effective trials
• More effective treatment
• Insights on pathophysiology
19
We need to recognize & forecast patient variation
Heterogeneity in lesion change in colorectal cancer
Nikodemiou et al. (2020)
20. • Make trials more informative
• ML to estimate treatment
effects, patient heterogeneity
• Automate capture of data
• Run more trials
• Deep integration with EHRs
and health systems
• Easier for trials to find patients,
easier for patients to find trials
• Adaptive trial design
20
We need to make clinical trials more powerful
ML to estimate individualized treatment effects
Zame et al. (2020)
21. RWD can make clinical trials more
powerful:
• Identify & quantify unmet health needs
• Forecast the demand for any therapy
• Subtype responders & non-responders
• Synthetic control arms
• Lower the bar for exploration, run trials
without running trials
Needs not just Big but Deep Data
21
We need to use Real World Data everywhere
Patients from historical
trials / RWE data
Inclusion /
exclusion criteria
Apply Propensity Score Matching
23. • Otherwise we risk repeating the
history of bioinformatics
• “We need someone to do our
data science”
• Needs a culture change
• Needs (willing) collaboration
• Needs investment
23
Deciding to use ML and data science is just the first step
24. • Labelled data is the new oil
• Unfortunately
• Data coverage is sparse
• Data is weird
• And also WEIRD
• Diverse data (more and unexploited
information)
• Lowers barrier to exploration
24
We need to invest in data
Reddy (2020)
25. How do we get this data?
• Usual RWE sources and …
• Real-time and intimate integration with
EHRs
• Devices
• Collaborate with national centres
• Long-term funding & broad
collaborations
25
We need to extend the reach of our data
27. • The odds are poor in drug development, but data science / ML / AI can
reduce the randomness
• We need methods that can cope with the complexity of real biological data
• Recognizing patient heterogeneity is a force multiplier
• Clinical trials can be made more powerful
• It’s an investment, not an overnight decision
27
Summary
28. Thanks
• Paul Metcalfe & ONC R&D ML&AI @AZ
• Irena Brookes-Smith & Health Informatics @AZ
• Krishna Bulusu & ECO @AZ
• Ray Lui, head ONC R&D Analytics
28
29. Confidentiality Notice
This file is private and may contain confidential and proprietary information. If you have received this file in error, please notify us and remove
it from your system and note that you must not copy, distribute or take any action in reliance on it. Any unauthorized use or disclosure of the
contents of this file is not permitted and may be unlawful. AstraZeneca PLC, 1 Francis Crick Avenue, Cambridge Biomedical Campus,
Cambridge, CB2 0AA, UK, T: +44(0)203 749 5000, www.astrazeneca.com
29
Confidentiality Notice
This file is private and may contain confidential and proprietary information. If you have received this file in error, please notify us and remove
it from your system and note that you must not copy, distribute or take any action in reliance on it. Any unauthorized use or disclosure of the
contents of this file is not permitted and may be unlawful. AstraZeneca PLC, 1 Francis Crick Avenue, Cambridge Biomedical Campus,
Cambridge, CB2 0AA, UK, T: +44(0)203 749 5000, www.astrazeneca.com
10
Notas del editor
With the cost of drug R&D having increase by ten-fold in the last 30 years and the success rate of a drug lead passing clinical trials has decreased by almost half to 12%, the calling to ending the gambling era of drug R&D is drawing near.