173 - The Evolving Role of Al in R&D (S12E8)

Welcome to the deep dive. Today we're diving
deep into artificial intelligence in research
and development, specifically how it's really
shaking things up in drug discovery and clinical
trials. Absolutely. It's not science fiction
anymore, is it? It's here now driving innovation.
Definitely not. So for this deep dive, we've
gathered quite a range of sources. We're talking
patents, pharmacokinetic studies, guides on anti
-cancer drug development, clinical trials. Right,
regulatory affairs info, even organic process
research literature. A real mix. Exactly. And
our mission really is to figure out how AI is
actually influencing each stage. from finding
targets right through to optimizing trials, and
maybe uncover some surprising bits along the
way. Sounds good. Where should we start? Okay,
let's unpack this. Starting at the beginning,
target identification and drug discovery. Right.
Traditionally, finding a good target. Well, it
took a long time. Sometimes it was just luck.
Yeah, serendipity played a big role. But now,
AI can chew through these massive data sets,
genomics, protein structures, disease info, much
faster. To pinpoint potential drug targets with...
Like, way more precision. Exactly. Much faster,
much more targeted. And we're seeing companies
built on this, like InnoPlexis. You mentioned
them. Yeah. InnoPlexis is a big global player.
They're heavily into AI for discovery and development.
And the patents prove it, right? Over 100 since
2017. Big filings in the US, China, EU, and importantly,
mostly in computer technology. That really tells
the story, doesn't it? It's about the algorithms,
the AI platform itself. They even have one called
OntiSight. OntiSight, yeah. And it's got ISO
certification for information security, which
is crucial given the sensitive data they handle.
That patent activity alone just screams commitment
to AI. It really does. And then there's Atomwise.
They developed their deep learning tech, AtomNet,
way back in 2012. AtomNet focuses on structure
-based discovery, right? Like virtual molecular
docking. Sort of, yeah. Using AI to predict how
small molecules will bind to a target based on
3D shape. On a massive scale. And they're collaborating
with Major Pharma Janssen, Eli Lilly, Sanofi.
Right. Biomotive, Hanso Pharma, too, using Adamet
to find targets and develop candidates, especially
on oncology, infectious disease, immunology.
Tough areas. And they even spun off a company,
X37. Yep. Back in 2019, focused on specific targets
in cancer, blood disorders, immunology. So AI
insights leading to focus development. OK, so
AI helps find targets and initial molecules.
What about screening those huge libraries? High
throughput screening, HTS. Right, HTS. testing
thousands, even millions of compounds quickly
to find those initial hits. AI makes this whole
process much more efficient. More efficient how?
Faster. Smarter analysis both really it's not
just about speed. It's about pulling meaningful
signals out of all that noisy data Identifying
patterns maybe a human wouldn't spot gotcha and
I saw something interesting in the sources about
combining AI with NMR nuclear magnetic resonance.
Oh, yeah, it's fascinating ligand based NMR methods
can actually screen compound libraries directly
even mixtures directly So you don't have to separate
everything first exactly you can identify binding
ligands just by their NMR signals skips a whole
laborious deconvolution step. That sounds like
a huge time saver. It absolutely is. A big boost
for efficiency in finding those early hits. OK,
so you've got hits. Now you need to understand
why they work. Structure -activity relationships
are. Right. How does the molecule structure dictate
its biological activity? AI and computational
methods are becoming indispensable here, often
paired with experimental techniques like NMR.
So NMR techniques like NOE. or STD, they can
show exactly which parts of the drug molecule
are touching the target protein. Precisely. Mapping
the binding epitope down to the atomic level,
almost. That's gold dust for medicinal chemists
trying to optimize the molecule. Making it more
potent, more selective. Exactly. And the sources
even mention things like spin labeling to find
other binding sites on the target, maybe for
drugs with more complex action. And all this
structural data feeds back into the AI. Yes,
that's the key synergy. The experimental data
trains the AI models, making them better at predicting
how structural tweaks will affect the drug's
properties. Really speeds up optimization. OK,
makes sense. So optimize structure. But then
you need to know how the drug actually behaves
in the body. Pharmacokinetics, PK. and ADME.
Right. Absorption, distribution, metabolism,
excretion. You absolutely have to predict these.
How does it get in? Where does it go? How's it
broken down? How does it get out? Critical stuff
for dosage, side effects, safety. Everything.
And AI is getting really good at helping predict
these ADME properties. Our sources mentioned
a specific liver model. The well -stirred venous
equilibrium model. It sounded complex. Yes. Yeah,
the name's a mouthful. But the core idea is about
how efficiently the liver clears a drug from
the blood. And it depends on only the free drug
getting into liver cells. Exactly. The drug not
bound to blood proteins. The formula you saw
just relates liver blood flow, how much the liver
extracts, that free fraction in blood, and the
liver's inherent ability to metabolize the drug,
its intrinsic clearance. And AI can help predict
all those different factors based on the drug
structure. It can certainly help model and predict
them, yes. And we also see molecular interaction
fields being used. What are those exactly? They
sort of map out how a molecule feels its environment,
its interactions, useful for predicting discovery
properties but also ADME. And understanding metabolism
is vital because of Drug interactions, right?
Like gemfibrozil, inhibiting that P450 enzyme.
Right, cytochrome P450 -2C9. Gemfibrozil is a
potent inhibitor. And other drugs like floxetine
or lancoprazole are metabolized by specific P450s.
You need to know this stuff. To avoid dangerous
combinations if someone's taking multiple meds.
Precisely. That's why studying drugs with intact
hepatocytes, liver cells in the lab is so important.
Because they have all the relevant enzymes. CYP3A4.
or CYP2C isoform? Exactly. It's the most complete
in vitro system. You feed that data into AI models,
you get a much better picture of real -world
metabolism. And this helps predict those drug
-drug interactions. I saw a rule of thumb. Ikey
ratio. Yeah, the inhibitor concentration over
the inhibition constant. If it's low, like under
0 .1, interaction's unlikely. Around 1, maybe.
Over 1, probably. And AI helps predict those
values, flagging potential issues before clinical
trials. That's the goal. Identify risks early.
Which leads us perfectly into preclinical development.
testing in labs, in animals, but for humans.
Right, and AI's predictive power for PK, ADME,
and even potential toxicity is hugely valuable
here, analyzing all that preclinical data. Helping
choose the best candidates to move forward. The
sources mentioned NCI cell line screens and those
human tumor xenograft models. Yeah, standard
tools for evaluating anti -cancer efficacy preclinically.
But it's all heavily regulated, right? FDA, ICH.
Absolutely. Preclinical talk studies have strict
guidelines. Ethical treatment of animals, data
reliability, it's non -negotiable. You're looking
for any sign of trouble before you dose a human.
Makes sense. So if a drug passes preclinical,
it's on to human clinical trials. Another place
AI is making inroads. Definitely. Optimizing
trial design is a key area, like calculating
sample size. Using AI to figure out the minimum
number of patients needed. Yeah, to get a statistically
meaningful result without exposing more people
than necessary. It's more ethical, more efficient.
And what about those adaptive trial designs?
They seem increasingly common. They are, especially
in Phase 2. Things like Pick the Winner, where
you focus resources on the most promising arms
early on. Or Seamless Phase 3 designs. Right.
Where you can roll straight into Phase 3 if the
Phase 2 data looks strong enough. AI could be
incredibly useful for analyzing the incoming
data in real time to make those adaptations robustly.
So AI helps analyze data during the trial to
adjust the design. Potentially, yes. Clinical
trial design is already this iterative process
between statisticians and doctors, often using
simulations beforehand. Computer simulations
to test different designs? Exactly. AI could
supercharge those simulations, explore way more
options, find more efficient designs. But you
still need that detailed protocol, drug, dose,
randomization, everything clearly laid out. Okay,
shifting gears slightly. What about actually
making the drug? Organic process research and
development. OPRND. Ah, yes, scaling up the chemistry.
Crucial, complex work. And maybe surprisingly,
AI is starting to impact this, too. How so? By
analyzing literature. Precisely. Think about
all the published papers, patents, reports on
chemical synthesis. AI can mine that data for
trends. Trends in reaction conditions. optimal
solvents, predicting scale -up problems. All
of the above. Imagine AI analyzing hundreds of
papers on, say, temperature screening in specific
reactors, or different solvent systems used for
similar reactions. Define the most robust, efficient
conditions for manufacturing this new drug. That's
the idea. Our sources even had examples like
finding a unified solvent system, one solvent
for multiple steps. AI could potentially help
identify those opportunities by analyzing vast
datasets. Streamlining the whole process. What
about crystallization? That seems critical for
the final drug form. Hugely critical. The solid
state affects dissolution, stability, formulation.
Everything. AI could analyze OPRND data on different
crystallization parameters, solvents, temperatures,
seeding. To predict the best way to consistently
get the right crystal form. Exactly. Optimizing
that final crucial step using data -driven insights.
Fascinating. And of course, all this happens
under the eye of regulators. Indeed. Navigating
regulatory approval is a huge hurdle. AI might
eventually assist in analyzing guidelines or
preparing submission documents. But the agencies
make the final call. Always. And good manufacturing
practices, GMP, are fundamental to ensure quality
and consistency in production. Right. And you
can't claim a drug is safe or effective before
it's actually approved by, say, the FDA. Absolutely
not. Promotion is strictly regulated pre -approval.
So looking ahead then, what are the really big
future potentials for AI in R &D? Oh, I think
we'll see even faster discovery timelines. AI
models will just keep getting better, integrating
more diverse data. And personalized medicine?
Yeah. Tailoring treatments? Yeah, definitely.
AI -driven patient stratification in trials,
figuring out who will respond best to which drug
based on their specific biology, genomics. That
sounds transformative. What about safety after
a drug is on the market? Another huge area. AI
analyzing real -world evidence, electronic health
records, adverse event reports, potentially spotting
safety signals much faster than we can now. Integrating
all these data types, genomics, proteomics, imaging,
clinical data. Yes. Getting that holistic view
of disease and drug response, that's where really
deep insights could come from. Which leads to
the big question, doesn't it? As AI keeps learning,
will it become the, I don't know, primary driver
of pharma innovation? It's a profound question.
What would that mean for medicine? It's certainly
a lot to think about. But what's clear now is
that AI is already impacting every stage. No
doubt. Accelerating discovery, optimizing preclinical
clinical stages, informing manufacturing. It's
pervasive. We definitely hit some aha moments
today. Those inoplexis patent numbers, the specific
atom -wise partnerships, even AI predicting optimal
crystallization conditions. Yeah, it gets down
to very specific applications. It's not just
a vague concept anymore. And it's important for
you, listening, to think about the wider implications.
Yeah. Right? The ethical considerations, the
practicality. Absolutely. How do we manage this
powerful technology responsibly? If you want
to dig deeper, maybe look into AI applications
in specific diseases, Alzheimer's maybe. Yeah.
Or how regulators are adapting. There's tons
out there. For sure. And it leaves us with that
final thought. As AI evolves, will it fundamentally
reshape not just drug discovery, but the very
future of healthcare delivery? A future that
seems to be arriving faster than we might think.