Session 1

Overview of AI and ML in Drug

Discovery and Development
https://www.forbes.com/sites/forbestechcouncil/2021/08/04/accelerating-
drug-discovery-with-mindful-ai/?sh=671ff6217aad 
AI drug discovery process provides an accelerated
pathway to the promise of personalized precision
medicines.
By exercising mindful AI, pharmaceutical companies can
improve the effectiveness of drugs in two important
ways:
• Getting drugs to market faster for the people who
need them.
• Treating more people by being more inclusive and
transparent across demographics and country borders,
thus striving for healthcare equity.
 FORMAT
40-50 Minutes Presentation
Review of Concepts, Papers, Companies and Tools
Followed by 10-20 Minute Break
Breaks Include Exercises Such as Reading, Youtube Clips or Problem Solving
Many but not all of the Breaks are Followed by 30 Minute Group Brainstorming
Sessions
Then Repeat
https://www.pharmaceutical-
technology.com/comment/artificial-intelligence-
future-drug-discovery/

THE BIG IDEA

More and more, big pharma is partnering with AI-
driven companies in hopes of more accurately
predicting drug candidates and cutting R&D costs
and time
https://www.pharmaceutical-
technology.com/comment/artificial-intelligence-
future-drug-discovery/

Our ability to make an impact on healthcare—

specifically on drug development—to benefit
people and society will be enlarged and enhanced
by AI in two ways:
1.  The timeline to bring new drugs to market will be
greatly shortened resulting in lower development
costs. AI will cut costs and reduce failure rates.
2.  With reduced development costs and accelerated
timelines for development, the ultimate cost to the
patient and healthcare system will decrease as well
while improving the benefit to patients globally.
https://asia.nikkei.com/Business/Pharmaceuticals/AI-
slashes-time-and-cost-of-drug-discovery-and-development

EXAMPLES:
1.  British AI startup BenevolentAI identified baricitinib, a drug
developed by Elli Lily for the treatment of rheumatoid arthritis,
as a potentially effective COVID-19 drug in just a few days. The
medication has been approved as a COVID-19 treatment in the
U.S. and Japan.

2.  Exscientia is attracting the attention of pharmaceutical and

biotechnology companies around the world. Evotec, a
German biomedical company, has jointly developed a new
cancer treatment with Exscientia. Human clinical trials of the
A2a receptor antagonist began in April, according to Evotec.
The candidate drug was discovered eight months after the
two companies launched the project.
https://spectrum.ieee.org/how-ibm-watson-
overpromised-and-underdelivered-on-ai-health-care

WHAT HAPPENED
In the eight years since, IBM has trumpeted many more
high-profile efforts to develop AI-powered medical
technology—many of which have fizzled, and a few of
which have failed spectacularly. The company spent
billions on acquisitions to bolster its internal efforts, but
insiders say this did not work.

IBM’s Watson venture into cancer drug discovery and

therapeutics resulted in a failed project with MD Anderson
Cancer Center. After 4 years of spiraling costs and missed
deadlines the project was terminated. The project was
originally contracted at $2.4 million and ended up costing
the Cancer Center $39 million.
https://www.linkedin.com/pulse/failures-some-ai-
approaches-drug-discovery-including-ibm-omar-
gadir/?trackingId=x7ogHTahQrybLng1u2vSZA%3D%3D

THE BIG IDEA

A good understanding of the data is a critical
prerequisite for attaining highly accurate results. This is a
big challenge as biomedical data is one of the most, if
not the most, complex area of data science.
- Extract In sights f rom Unstructured Data
- Pay attention to nomenclature
- Clean the data first
- Seek > 90% accuracy
- Understand biomedical data first (use domain experts)
 <
mber
nu
AI and Big Data for R&D Drug Discovery
m
An overly simplified view of the process of
be
r>
developing new molecular-targeted therapeutics,
'from gene to drug’
Areas
where BIG
Areas
DATA is
where AI
abundant
is being
used

Collins & Workman; Nature Chemical Biology 2, 689 -

 <
mber
nu
Why do we care where the Big Data is?
m
be Today, AI is mostly Machine Learning
r> over Big Data
DO NOT ignore traditional AI:

Multiple of methods:
Heuristic search over large spaces

Deep learning – useful when data is

Rule based systems (or expert systems)
quite dense, often not the case in drug
discovery
Natural language extraction from text

Probabilistic reasoning and fuzzy logic

Machine vision (for interpreting imagery)

Decision trees

Clustering

Supervised vs. Unsupervised Learning

Garbage In = Garbage Out, It’s All About

the Data
 <
mber
nu
“Big Data” tools for
m
be
toxicology
r> Best case scenario: Big Data is available; there is a clinical need; and
AI algorithms work

Predicting toxicity of molecules in-silico

Using experimental data from HTS

Millions of compounds in hundreds of assays

“Dirty” data, voluminous amounts, challenging to analyze

Tox21, ToxCast in the EPA

Pathway modeling and individual molecules predictions

From molecules to populations

Machine learning and genetic algorithms were the best tools for these
sets
https://www.epa.gov/sites/production/files/2013-
20 Minute Break
Read the Articles Linked in the Slides
What metrics can AI improve in DDD?
> Shorter Development Time
> Decreased Failure Rate
> Greater Novelty, Better Starting Points
> Lower Cost

How can this affect the $2.6B cost to develop a drug?

Does it result in more time on patent?
Will there be more "best in class" or "first in class" drugs?
Will the drug paradigm change f rom targeted
therapeutics to network biology?
https://papers.nips.cc/paper/2012/file/c399862d3b9d6b76c8436e924a68c45b-Paper.pdf

--> Go visit this paper

--> Supervised Deep Learning
--> Pay Attention to Figure 2
--> Read Discussion on Overfitting
https://www.sciencedirect.com/science/article/pii/S1359644617303598

Highlights
•Deep learning technology has gained remarkable success.
•Shows recent applications of deep learning in drug discovery research.
•Some popular deep learning architectures are introduced in the current study.
•Future development of deep learning in drug discovery is discussed.
https://www.linkedin.com/pulse/deep-learning-drug-
discovery-new-frontier-oscar-bastidas/?
trackingId=qjsgI9S3T9ewRSAVWb3jFA%3D%3D
One of the most appealing abilities that DeepNetts® has to
contribute is its freedom from quantitative structure activity
relationship (QSAR) input as an obligatory predictor variable.
Although extremely useful in considering molecular
characteristics as part of a data analytics campaign, there are
situations where obtaining structural information such as in
the form of the simplified molecular-input line-entry system
(SMILES) is extremely challenging. This is of particular concern
especially for thousands of molecule data points that might
not be registered in databases such as ChEMBL (or the lab
might have them listed by more esoteric names not used in
such databases). In such cases, being able to carry out a deep
learning analysis of the data using physical and chemical
properties already provided as part of the compound library,
can prove an invaluable supplement or even alternative to
QSAR. 
Tensorflow PyTorch
https://github.com/tensorflow/tensorflow    https://pytorch.org/ 
For Deep Learning Easy to Use
https://pytorch.org/tutorials/ 
Online Tutorials
May be used on AWS
Makes use of a Generative Adversarial Network
GAN

See "Generative Deep Learning" by David Foster

A GAN is a battle between two adversaries, the

"generator" and the "discriminator".  The generator
tries to convert random noise into observations and
the discriminator accepts or rejects them.
 https://insilico.com/  
 20 Minute Break
This Youtube will teach you Deep Learning in 10 Hours
Bookmark it if you are new to the subject
https://www.youtube.com/watch?v=JMxbypF825w

During the break, read the following slide

“Trends and Developments in Artificial Intelligence: Challenges to the Intellectual Property Rights Framework” prepared for the European Commission.

• First, artificial intelligence has been applied to finding molecular targets, being a molecular structure with which an active pharmaceutical ingredient interacts
and thus triggers its effects. Most drug targets are proteins such as receptors, enzymes, transporters and ion channels. Nucleic acids and structural proteins can
also be targets. Scientists have recently developed a machine learning algorithm capable of predicting biological targets of prospective drug molecules. An
example is the Bayesian analysis for the determination of drug interaction targets (BANDIT), which is designed to use virtually all available data on any potential
drug to predict which enzyme or receptor or other target in the cells it interacts with to achieve its therapeutic effect. BANDIT can streamline the process of target
identification by limiting the possibilities to perhaps only one or two targets, which can then be studied.

• Second, artificial intelligence has been applied to finding a “hit” or “lead”. Automated high-throughput screening is a method for finding so-called hits and thus
candidates for lead structures. A lead is a chemical substance that is investigated in the course of drug design as a starting point for the development of a drug
candidate. A lead already shows a desired biological effect in vitro, but does not yet have the qualities required for a drug substance in terms of potency, selectivity
and pharmacokinetics. A lead differs from any hit from high-throughput screening in that its structure allows the synthesis of analogue molecules. A lead can also
be a natural substance whose pharmacological properties must be improved by chemical modifications.

• Third, artificial intelligence has been used in drug repurposing. Many drugs are small, flexible molecules that not only interact with the originally determined
target structure, but can also interact with other biological structures. Such associated effects can lead to the identification of new applications.

• Fourth, artificial intelligence can help in identifying candidate substances by screening existing databases. Scientists have applied machine learning models to
identify completely new antibiotically-active substances. The machine learning models can analyse molecular structures and correlate them with certain
properties, such as the ability to kill bacteria, and convert them into algorithms.

CONTINUED....
• Fifth, artificial intelligence has been used in polypharmacology. Drug development in recent decades has been guided by the paradigm “one drug – one target
structure – one disease”. This principle is based on the approach that a pharmacological agent is used to selectively address a target structure in the body in order
to treat a specific disease pattern. A large number of serious illnesses, such as cancer, mental disorders or cardiovascular diseases, often cannot be explained by a
single biological defect, but only the interplay of complex and usually networked malfunctions that together produce the clinical picture. Artificial intelligence can
assist.

• Sixth, artificial intelligence is a useful tool in the development of vaccines. Artificial intelligence can help with gene sequencing, so as to decode the genetic
material of the virus. Artificial intelligence can also help with the simulation of vaccines. In a simulation, possible vaccines can be “released” on the virus. The data
on the nature of the virus and the vaccine may be sufficient to simulate how they react to each other. This makes it possible to exclude less promising vaccines
even before the actual tests begin.

• Seventh, artificial intelligence can assist in determining the three-dimensional structure of proteins including their folding forms. The Alphabet subsidiary
DeepMind used deep learning, a subfield of machine learning, to develop the AlphaFold system, which generates 3D models of proteins based on their genetic
sequences. AlphaFold has been used to predict protein structures of the SARS-CoV-2 virus. This is important in research for a vaccine. The antibodies of a vaccine
must act precisely on the proteins of the virus in order to neutralise it.

• Eighth, let me say something about databases. There are many databases that are available to integrate diverse information of molecular pathways, crystal
structures, binding affinities, drug targets, disease relevance, chemical properties and biological activities. Artificial intelligence could be used to harvest
information from these databases to design new compounds both for single targets but also in polypharmacology.

• Ninth, artificial intelligence can also be used to support clinical trials by detecting the disease in patients, identifying the gene targets and predicting the effect of
the designed molecule and the on- and off-target effects. Further, patient selection is a critical process in clinical trials. In clinical studies, the relationship between
human biomarkers and in vitro phenotypes is essential. It allows a more predictable, quantifiable evaluation of the uncertainty of therapeutic responses in a given
patient. The development of artificial intelligence approaches for the identification and prediction of human relevant disease biomarkers enables the recruitment
of a specific patient population in Phase II and Phase III clinical trials. The success rate in clinical trials can thus be increased by predictive modelling using artificial
intelligence in the selection of a patient population.
What new problems in drug discovery and
development can we apply Machine Learning to to
save time or money, or to increase value, or decrease
failure rate?  How?  What data sources could be
used?

Group Brainstorming Best Practices

- All Brains are Equal
- Warm Up with an Easy Problem
- No Criticism
- Allow Whackiness
- No Ownership of Ideas
- Everyone Participates
- Keep it Short
 Session 2

Bayesian Reasoning in Drug

Discovery and Development
 20 Minute Break
Peruse the Bayesian PDF During the Break
https://www.nature.com/articles/s41467-019-12928-
6 

"BANDIT" Uses an integrative Approach

- Predicts Target Interactions
- Identifies Specific mechanisms
- "Walkthrough:.....Visuals in Paper
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3607962/

Most Ai has focused on energy binding

Toxicity and Bioavailability are the big preclinical
problem.  Bayesian Reasoning can bridge the gap.
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3657986/

Bayesian statistics provide a formal mathematical

method for combining prior information with
current information at the design stage, during the
conduct of the trial, and at the analysis stage. It is
easier to implement adaptive trial designs using
Bayesian methods than f requentist methods.
Bayesian designs provide an efficient and effective
method for evaluating new molecules during the
early phases of drug development. The Bayesian
approach can also be applied for postmarketing
surveillance purposes and in meta-analysis. The
basic tenets of good trial design are the same for
both Bayesian and f requentist trials. 
FDA’s approval of a drug for a medical indication is
based on the totality of the evidence for
efficacy and
safety f rom clinical studies. Disapprovals can https://hbiostat.org/doc/bayes/whybayes.pdf
sometimes be revisited but once a
drug is approved for
market the decision is seldom revisited unless
significant evidence for safety
problems emerges.
Consider late-phase studies, and for now pretend that
efficacy and safety
can be assessed independently. In
reality, regulators consider benefit-risk tradeoffs
whereby the
amount of efficacy that must be
demonstrated is increased when there is a safety
problem, and
the amount of harm that can be
tolerated is increased when the clinical benefit is
substantial. The Bayesian approach can formalize the
analysis of benefit-risk tradeoffs.
https://www.berghealth.com/
Heavy User fo Bayesian Approaches

BERG is a clinical-stage, AI-powered biotechnology

company taking a bold ‘Back to Biology™’ approach
to healthcare. By leveraging our proprietary
intelligence platform – Interrogative Biology®, our
goal is to map disease and revolutionize treatments
for patients around the world.
 Bayesian Reasoning
Brainstorming
What problems can be solved using Bayesian Reasoning in DDD?
 Session 3

Natural Language Processing 

in Drug Discovery and
Development
 20 Minute Break
Watch this Youtube: 
https://www.youtube.com/watch?v=5ctbvkAMQO4
https://www.bio-itworld.com/news/2021/07/19/why-nlp-is-the-
key-to-interpreting-unstructured-data-in-pharma-drug-
discovery
https://www.drugdiscoverytrends.com/whats-driving-
the-natural-language-processing-revolution-in-
pharma-and-life-sciences/

In discovery, accessing literature sources for landscapes of gene-

disease associations reduces manual curation and presents
clusters of potential diseases for any chosen gene to enable
indication targeting.
Also, in discovery, developing an integrated systems pharmacology
approach allows reliable prediction of potential ‘on-target’ and ‘off-
target’ Adverse Drug Reactions (ADRs) for new drugs in
development.In safety, using NLP to enhance the searchability of
internal preclinical toxicology safety reports.
In development, many applications in clinical trial analytics, for trial
design, meta-analysis, and clinical competitive intelligence.In the
real world, extracting epidemiology metrics and models from the
literature.
In medical affairs, capturing internal and external intelligence
streams around product brands into one integrated environment
for visual analytics that aid brand team decisions.
Also, in the post-market, looking across scientific literature and
prescription databases to enable a better understanding of drug-
drug interactions and co-prescription trends.
https://www.mmitnetwork.com/member-content/natural-language-processing-has-promise-in-drug-discovery-patient-
identification/ 

✦ Novo Nordisk uses NLP-based text mining to better understand the concerns that patients and providers have about the use
of its products. The company looks for data on such topics as safety, efficacy, pharmacokinetic/pharmacodynamics modeling,
randomized controlled trials, patient populations, dosing and devices. According to Novo Nordisk, use of NLP for data extraction
has replaced scanning of manual real-world data sources, allowing for the elimination of two full-time equivalent positions and
automating the process of generating insights.
✦ GlaxoSmithKline plc drug safety teams use NLP text mining to streamline the literature review process for pharmacovigilance
and ensure better patient safety. They also use NLP search tools to review medical literature for safety signals. In one project, the
company found an average of 60 new adverse event references per day over a 300-day period that included a review of 11,000
documents and 20 marketed products. The company calculated that similar searches, if performed manually, would have taken
thousands of hours and included less precise findings.
✦ Sanofi also used NLP to help identify linkages between specific genes, diseases and drug sensitivity for a specific multiple
sclerosis biomarker project. Using NLP, the company’s search identified all of the 22 previously published autoimmune diseases
and drug sensitivities associated with HLA alleles and haplotypes, but also uncovered another 33 novel unpublished disease and
drug-sensitivity associations.
https://www.linguamatics.com/solutions/drug-
discovery 
See Case Studies for:
Target Prioritization
Target Selection
Text Extraction
https://www.outsourcing-
pharma.com/Article/2019/01/30/Sanofi-NLP-text-
mining-and-other-AI-tech-to-advance-findings-
improve-data-use

Sanofi: NLP text mining and other AI tech to

advance findings, improve data use30-Jan-2019 By
Melissa FassbenderReal world evidence extraction is
one area where Sanofi is using NLP text mining, a
technology with applications across the R&D
pipeline, f rom target identification to clinical trial
design and pharmacovigilance.
NLP Brainstorm
What are applications of NLP in DDD?
 5 Minute Break
Bookmark this 5 hour tutorial on NLP if you want to learn the basics
https://www.youtube.com/watch?v=68lIfswwG2A
 Session 4

Target Discovery in Drug

Discovery and Development
Ways Big Pharma is Applying
AI to Target Discovery
https://www.biopharmatrend.com/post/34-biopharmas-hunt-
for-artificial-intelligence-who-does-what/
REVIEW ARTICLE AND DISCUSS TOGETHER

 AI Target Discovery Survey

20 Minute Break
How is your company applying AI to target discovery?  Research and share only
nonconfidential information.  Discussion to continue after the break,
“Attention-based neural network system” (basically a
neural network that can focus on specific inputs in
order to increase efficiency). It’s able to continually
refine its own predictive graph of possible protein
folding outcomes based on their folding history, and
provide highly accurate predictions as a result.

https://techcrunch-
com.cdn.ampproject.org/c/s/techcrunch.com/2020/11/30/alphabets-
deepmind-achieves-historic-new-milestone-in-ai-
based-protein-structure-prediction/amp/
The DeepMind database consists of 350,000
newly-predicted protein structures and the
company plans to release an additional 100
million protein structures over the next few
months – as Demis Hassabis, the CEO and Co-
founder of DeepMind puts it: “Almost every
sequenced protein known to science – over 100
million structures … a veritable protein almanac
of the world.” This massive database has been
made available in a searchable format and the
best part? It’s completely f ree!
https://www.longevity.technology/3d-structures-
of-all-protein-youll-have-to-google-that/
In Session Discussion
How can this resource be applied now?
Target Identification deals with identification of function of possible molecular target (genes/proteins of a small molecule) and its
role in a disease, which is aimed at finding the efficacy target of a drug. This requires evaluation of functional genomics,
structural genomics, proteomics, and cell based assays (in-vitro), animal research (in-vivo) assays.AI is being used to analyse
Drug Information Bank (contains drug candidates, gene expressions, protein-protein interactions and clinical data records) from
a public library for predicting therapeutic potential. For example, applying feature engineering by deep autoencoder, relief
algorithm and binary classification by Xgboost algorithm on ‘genome-wide protein interaction network, drugs and their targets
information’ to produce scores for potential targets to enable target prioritization.For finding of drug target site, discrete
chemicals can be encoded into continuous latent vector space, which permits gradient-based optimization in molecular space,
and allows predictions using graph convolutional network based on binding affinity and others properties.In the effort to
understand the detailed spatial 3D structure of proteins and molecular complexes, AI platform also rely on training computer
vision and machine learning models on cryo-EM microscope data (2D structure).Selection of drug candidates requires series of
desired properties especially pharmacokinetics, pharmacological and toxicity profile. AI drug design algorithm Reinforcement
Learning can be successfully deployed for a ‘Simplified molecular input line-entry system’string, which is a specification in the
form of line notation for describing Chemical species using short ASCII strings. It includes potential energy measurements,
molecular graphs with varying weights for atoms or bonds, coulomb matrices, molecular fragments or bonds, atomic
coordinates in 3D etc.
https://www.wipro.com/holmes/4-application-areas-of-artificial-intelligence-in-drug-discovery/
Tuesday, June 22, 2021 - Insilico
Medicine is pleased to announce that it
has entered into a collaboration with
Teva Branded Pharmaceutical Products
R&D, Inc. to utilize Insilico's generative
machine learning technology and
proprietary PandaOmics Drug Discovery
Platform, which aims at identifying
novel therapeutic targets implicated in
a variety of diseases.
 Session 5

Lead Discovery in AI Drug

Discovery and Development
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7577280/ 
20 Minutes Review in Session
Outstanding overview
Presents a Dozen Tools
https://pubs.acs.org/doi/10.1021/acs.molpharmaceut.8b00930 
Review the paper
Cloud Pharmaceuticals
Known for AI Based Drug lead Design
See Approach on Next 2 Charts
 20 Minute Break
Read the Article in the Next Chart
And also: https://pharmaphorum.com/news/gsk-ai-drug-design-cloud-
pharmaceuticals/
https://www.pharma-iq.com/pre-clinical-
discovery-and-development/articles/an-effective-
way-to-apply-ai-to-the-design-of-new-drug-lead-
compounds
REVIEW THIS ARTICLE DURING THE BREAK
WE WILL ENGAGE IN A DEEP DISCUSSION

You may also be interested in this paper:

https://ieeexplore.ieee.org/document/8409367
https://insilico.com/
See website
 30 Minute Break for
20 Minute Exercise
View: https://www.youtube.com/watch?v=hY9Bc3mtphs
https://www.future-science.com/doi/10.4155/f mc-
2018-0212 
 “A molecule that can interact with a protein target
can usually interact with upward of 300 proteins,”
Cyclica’s CEO Naheed Kurji says. “If you’re designing
a molecule, it behooves you to consider the other
299 interactions that could have disastrous effects in
humans.” 
Cyclica is attempting to find individual compounds
that can interact with dozens of target proteins yet
avoid interacting with hundreds of other proteins,
Kurji explains.
Open Drug Discovery ToolkitOpen Drug Discovery Toolkit (ODDT) is a modular and comprehensive toolkit
written in Python for use in cheminformatics, molecular modeling. It does not take advantage of a well-
known deep learning library and it requires only SK-learn.ODDT includes many state-of-the-art methods, like
machine learning scoring functions (RF-Score and NNScore). The goal is to create a tool that other people
can customize for their computer-aided drug discovery. Apart f rom the tutorial linked in the repository,
there is a reference paper published in the Journal of Cheminformatics. To manage and translate molecular
data it requires installed the already mentioned RDkit or Openababel.As interesting features, it includes
3 Molecular shape comparison methods (USR, USRCAT, and Electroshape)
https://www.researchgate.net/figure/List-of-AI-
Based-Computational-Tools-for-Drug-
Discovery_tbl1_334475203 
17 Tools - Walkthrough
 Session 6

Survey of Companies in
AI Drug Discovery
 230 Companies ??

https://blog.benchsci.com/startups-using-artificial-intelligence-in-drug-discovery

Summary Walkthrough Discussing Select Companies

 Open Discussion
Business Models for Collaborations - AI Vendors and Big Pharma
 More Well Funded
Companies
https://www.nanalyze.com/2021/04/companies-ai-drug-discovery/
10 Minute Break
No Exercise
 Session 7

Repurposing Drugs with AI

https://blog.benchsci.com/startups-using-artificial-intelligence-in-drug-
discovery#repurpose_existing_drugs  
https://www.thelancet.com/journals/landig/article/PIIS2589-7500(20)30192-8/fulltext 
 20 Minute Break
Review Repurposing Review paper, then Discuss
https://www.sciencedirect.com/science/article/pii/S2590257121000298
Market Interest
How it Changed During Covid
Cocktails and Precision Medicine
Fast to Clinic
Business Models
 Open Discussion
Repurposing of "Drug Cocktails" with Biomarker
 Session 8

AI in Preclinical Research,
Biomarker Discovery and
Pipeline Management
Can we:
1.  Reduce time?
2.  Reduce cost?
3.  Fail early?
4.  Pick the best compounds?
5.  Pick the best programs?
6.  Increase success rate?
https://atomscience.org/
Industry Consortium
Atom is a partnership for collecting data for AI in drug discovery, particularly around problems with Pk and toxicity.
We will explore the Atom partnership and consortium.
Where does the data come
from?
DISCUSSION
https://www.outsourcing-
pharma.com/Article/2015/04/09/Predicting-oral-
bioavailability-is-better-with-AI-says-
PharmaInformatic
 Preclinical CROs Using AI
https://electronichealthreporter.com/how-preclinical-cros-are-now-
using-future-technology-to-conduct-health-research/

DISCUSSION
https://blog.benchsci.com/startups-using-
artificial-intelligence-in-drug-
discovery#design_preclinical_experiments 
-
https://blog.benchsci.com/startups-using-
artificial-intelligence-in-drug-
discovery#validate_and_optimize_drug_candidates
 20 Minute Break
Download and Read the White Paper During break - to warm up
with AI in Biomarkers
https://www.iqvia.com/library/white-papers/breaking-new-ground-with-digital-
biomarkers 
To be discussed
https://www.innoplexus.com/artificial-
intelligence-driven-biomarker-discovery/
41% of New Drug Approvals are Associated  with
Biomarkers 
Pay attention to the data sources used
https://www.innoplexus.com/technology/ 
Technologies Used
AI for Biomarker Discovery
an Active Research Topic
https://www.frontiersin.org/research-topics/18505/artificial-
intelligence-a-step-forward-in-biomarker-discovery-and-
integration-towards-improved-cance 
An Example Application:  Dermatitis  https://www.pnas.org/content/117/52/33474
https://www.sciencedirect.com/science/article/pii/S2405471221002507
 Biomarkers with AI for
Alzheimer's
https://www.technologynetworks.com/neuroscience/news/using-
advanced-ai-to-discover-diagnostic-biomarkers-and-drug-
targets-for-alzheimers-341638
https://www.tandfonline.com/doi/full/10.1080/17460441.2021.1918096
AI for Pipeline Management
What programs should I go forward with?  Which ones should fail
early?  AI can help....
https://www.iqvia.com/solutions/commercialization/brand-
strategy-and-management/product-pipeline
https://www.bloomberg.com/news/articles/2021-
02-05/out-of-grief-mit-s-andrew-lo-invented-a-
better-way-to-finance-biomedical-innovation
-
ROI on Pipeline is 12%, but beta is enormous.  Need
150 programs (in oncology) to reduce the variance of
program success to allow financing by the bond
market.
Possible error in assumptions relate to correlation of
programs
This would require $30B.
How can AI do better?
Active Research...not fully solved
 20 Minute Break
Brainstorm on Application of AI to Pipeline Management
Session 9

AI for Clinical Trials 

https://www.clinicalleader.com/doc/how-
machine-learning-will-eliminate-failures-in-
clinical-trials-0001 
One of the hottest topics in drug development is the
use of AI in clinical trials. AI is defined as a science
concerned with building smart machines capable of
performing tasks that typically require human
intelligence. 
continued....
Now imagine this scenario: You are developing a treatment for pancreatic cancer. You are able to use a machine
learning solution to determine whether a molecule might be a good candidate for that disease.  It could also be
used to determine if the number of patients you require for the study is attainable. Next it will tell you which cancer
patients would be the best fit for your product, based on personal characteristics and characteristics of their tumor.
You can also eliminate those patients who will likely not benefit f rom the treatment. Incidence of disease is not the
same in all areas of the world, so you can determine which countries will most likely provide access to the patients
you require. Once you identify the right countries in which to perform your trial, you can also identify which clinical
sites in that country will be most successful at recruiting the patients you require. You could also determine which
CRO would be the best fit for your study, based on global presence, expertise, and past performance. And best of
all, that required information could be gleaned by using hard data, not by human guesswork.
Making Go /
No-Go Decisions

The biggest reason a drug program

fails is a failed Phase 3 trial
 Opportunity: Use AI to accelerate and de-risk drug
development
Generate
experimental data
At each stage of the funnel, evaluate all
leads in parallel with FDA approved
Train algorithms
drugs and failed drugs
Train machine learning algorithms on
public and experimental data with clinical
Learn markers of success or failure as the target
clinical success Models will learn markers of clinical
success from pre-clinical data
Generate
predictions Generate predictions of clinical success for
the leads
Focus capital on
the best leads
At each stage of the funnel, all
leads are competing to move
Design of
Clinical
Trials With
AI
13.8% Success Rate
for Drugs thru 3
Phases of Clinical
Trials
 Enhance Adherence Using Wearables and
Sensors

Better Endpoints by Analyzing Data Across

AI for trail Multiple Trials

Management
More Accurate Assessment of Variance of Patient
Data Results in More Optimal Trial Design –
Reduce Cost by Including the Right Number of
Patients
Detection of Signals in Data That Humans
Cannot – to Better Assess Efficacy

Assess Protocol Feasibility Using Real World Data

(RWD)

Assess Site Performance With Real Time

Monitoring
Enrollment of Clinical
Trials Using AI and EMR,
Genomics

https://deep6.ai/
 Search EMR for Matching Patients at
Ways AI Potential Sites
Enhances NLP and Ontologies for
Clinical Extracting Info f rom EMR
Trial AI Based Advertising on Online
Enrollment Health and Wellness Sites

AI Based Advertising at Ancestry and

Other Genomic Testing Services

Better Screening Criteria Based on

Data Mining
Using AI for Better
Management and Delivery
of Trials

Better management of
multisite trials
 ASSESS SITE REAL TIME AUTOMATE

AI for PERFORMANCE
(ENROLLMENTS,
DROPOUTS,
COMPLIANCE)
ANALYSIS OF
DIGITAL
BIOMARKERS
SHARING ACROSS
MULTIPLE SITES
[DOES NOT
REQUIRE AI]

Trial
Conduct SMART PHONE
ALERTS, E-
ALERTS FOR
MISSED VISITS,
TRACKING OF NONCOMPLIANCE
MEDICATION
COMPLIANCE
 A Phase 3 failure is typically the end for a drug
program
Can We
Save a But what if a drug is safe and efficacious for a
subset of the population but not the entire
Drug if population?
Phase 3 Ai can analyze clinical data and EMR / Genomics
and determine if a drug can be saved for a portion
Fails? of the population
If this can be effectively done, the cost of drug
development goes down by eliminating a total
failure and providing a product for some people
Not yet proven, but a worthy exploration
 Session 10

AI for Drug Safety, Drug

Interactions, and EMR Mining
https://www.rtinsights.com/how-ai-and-machine-learning-can-
revolutionize-drug-safety-monitoring/ 
Approximately 80% of healthcare data resides in unstructured
formats like emails and paper documents. Today, unstructured
data on AEs needs to be aggregated and correlated from disparate
and expansive data sources, including social media, email, online
communities, and other digital formats. Currently, most PV and
safety professionals need to review AE data by going through
inefficient manual processes. This can slow down clinical trials and
potentially delay the delivery of new medications to patients.
The introduction of artificial intelligence (AI), machine learning
(ML), and natural language processing (NLP) tools to the life
sciences industry is making it possible to take these large,
unstructured data sets and turn them into actionable insights at
unprecedented speeds. Investment in AI and ML in healthcare
sciences is expected to grow to $8 billion by 2022. This investment
signals a dramatic increase in the quantity and quality of drugs on
the market, fueled by a need for faster time to market for new drug
therapies and a new approach to PV.
cont
Speed literature searches for relevant information
Scan social media across the globe to pinpoint AEs
Listen and absorb audio calls (e.g., into a call center) for mentions of a company or drug
Translate large amounts of information f rom one language into another
Transform scanned documents on AEs into actionable information
Read and interpret case narratives with minimal human guidance
Determine whether any patterns in adverse reaction data are providing new, previously unrealized information
that could improve patient safety
Automate case follow-ups to verify information and capture any missing data
Papers and Abstracts: AI
and Drug Safety
https://pubmed.ncbi.nlm.nih.gov/31383376/
10 Minute Break - Review Abstracts
Session 11

Future Directions
Targeted therapeutics are on-Off Switches
Modulated Biology Allows a New Paradigm
Optimal Controls of Pathways
Requires High Fidelity Systems Biology
AI Will Make a Difference Here
Draw f rom Literature Extraction, Pathways
Databases (ie KEGG), Protein Databases (ie
DeepMind)
Forge Network Model
Biochemical Reaction Chemistry forms Basis of
Dynamic Models
AI for System Identification
Drugs are Polypharmacy with multiple releases or
delivery mechanisms
Open Discussion
Future Directions of AI in DDD
Map data ©2021 Google