De-Risking What Matters

In this episode of Innovation to Impact: Ruminations & Ramblings, Szczepan Baran, Brian Berridge, and Nick Kelley tackle one of the biggest problems in modern drug development: we keep adding more technology, more data, and more complexity, yet clinical attrition remains painfully high. Across discussions on AI, NAMs, digital biomarkers, animal models, translational science, and organizational culture, they argue that innovation fails when tools become the strategy instead of serving clearly defined patient-centered decisions. The conversation explores why reverse translation matters, how AI should function as a decision-support system rather than a magic oracle, why “decision warranties” may become essential in AI-enabled science, and how the industry continues to confuse activity with progress. This is a candid, often uncomfortable discussion about predictivity, accountability, translational learning, and what it would actually take to build a drug development system optimized for patient outcomes instead of platform hype. 

What is Innovation to Impact: Drug Development, AI, and Regulatory Strategy?

Innovation to Impact is a podcast on decision-grade drug development in regulated environments.

We examine how high-stakes go/no-go calls are made inside pharma and biotech, and what evidence is required for new tools to change those decisions without creating hidden risk.

Each episode focuses on predictivity, translational risk, decision rights, and accountability (what breaks, who owns it, and what triggers a stop).

This is not a podcast about technology trends. It is about disciplined innovation that can survive audit, scale, and real-world biology.

Hi, everybody.

I'm here for another session of Innovation

to Impact with my good buddies,

Nick and Shepon,

where we get together and go to bounce

around ideas and talk about issues that

we've experienced and seen in the

challenges related to adopting and

adapting innovation and technology to the

challenges of drug development.

So if you've watched one of these before,

you know that we have

some interesting conversations and we all

come at it from a bit of a

different perspective.

And so it's kind of an interesting

opportunity for us.

Certainly entertaining for us and

hopefully entertaining for you all as

well.

Before we launch into this particular

session,

I'm interested in what's happening in your

world, Jess.

Well, so, I mean,

I can jump in first.

Today, for me,

this is going to be an easy one.

I just started a new gig at NSTEM

as a chief scientific officer.

So that's certainly pretty exciting for

me.

Looking forward to that role and working

with everybody there and also other

partners and collaborators.

On the other side also,

the weather for us, you know,

going from like,

forty degrees all the way to AD, snow,

no snow the next day.

It's been a lot of fun as well.

So how about you, Nick?

No, I mean,

as you guys were pointing out, I mean,

what's new in my world is my first

week off crutches after a bit of a

sports mishap in my Achilles some while

back.

And, you know,

what's relevant today is it really makes

you think about, you know,

what does health mean?

What does carrying, you know,

extending your quality of life into later

years mean?

And

what can it look like if you don't

because you know certainly you know it's a

whole new level of empathy for people that

have challenges kind of you know getting

around it um or any disabilities of any

sort right so you know how much more

effort it took to get through a day

daily routine um you know and also with

you know just in time to make it

to Paris for a conference more on a

investor startup sort of sort of Summit um

you know the interesting themes there one

you know, it's a lot around, you know,

cloud code and how, you know,

the barriers to software, you know,

engineering and writing are coming down

and what is that gonna mean for,

you know, drug discovery and development,

and also how easily are, you know,

all of us going to be automated away

by, you know,

like software engineering may or may not

be.

And, you know, some of the consensus is,

you know,

know like some of the more resilient areas

will be highly complex innovative places

like biomedical research but you know

you'll have to be able to adopt and

leverage these tools and then you know it

kind of fits into the theme of some

of what we were talking about this month

in terms of you know how to you

know increase let's say the uh

effectiveness efficiency de-risk um

getting new medicines to market.

So I think from that point of view,

the world fits together in a lot of

ways.

Yeah,

the common theme there is the world's not

getting any simpler, is it?

And it's an interesting amalgamation of

challenges and opportunity.

And for me personally,

like a lot of us,

I'm spending a lot of time engaging our

healthcare system because I have an aged

parent

who is increasingly needing care.

And so it's an interesting exposure to the

clinical end of where I've spent most of

my career on the preclinical side.

And so even though I'm sort of an

aficionado of the translation from

preclinical to clinical,

just getting those kinds of insights into

the way clinical medicine works and how it

engages patients is an eye opener.

not only for how it affects my mom

in this particular case,

but thinking about my long-term trajectory

from a health perspective and thinking

about,

I need to do a better job of

trying to make sure that I have to

rely on the healthcare system as little as

possible.

That said, I have, as you all have,

had a lot of satisfaction out of

working in a field that is primarily

focused on trying to do things that help

patients.

In particular for us,

trying to link all these opportunities

that we increasingly have to do those

great things for patients.

So with that,

just kind of evolving into the

conversation that we wanted to have today.

And it's kind of reflecting on what we

recently wrote in a common theme.

And that common theme is the

quantitativeness of the work that we do,

particularly on the preclinical side and

even extending into clinical medicine and

how the numbers relate to patients and

those kinds of things.

And as a pathologist who's worked in

preclinical safety much of my career,

you know,

I've always been kind of

intrigued,

interested in the math problem that is

drug development.

And that is the fact that we identify

novel drug targets and develop novel

molecules that we think will do great

things for patients at some level of

exposure.

And as we go through development,

you find out that you actually need more

of it on board than you thought you

needed to get the benefit.

But at the same time,

the safety challenges are sort of nibbling

away at that at that margin,

that that opportunity for benefit.

And so, you know,

the end result of getting a marketable

drug has everything to do with whether you

can maintain a margin of safety versus

benefit that, you know,

allows the patient to to get the good

bits without having to suffer from the bad

stuff.

And again, as a pathologist,

I realize that with the computational

capabilities that we've got,

the AI opportunities that we have,

that a lot of the data that I've

generated has been much more,

it's morphologic and it's qualitative.

I give the changes that I see in

the tissues that we examine big fancy

words and make some qualitative assessment

about how bad they are.

that kind of data doesn't always fit into

the kinds of models and technologies that

we have.

And so I think a lot about, well,

how do we make what I do more

quantitative so it's more integratable

with other parts of our business and

actually we can take better advantage of

the kinds of technology opportunities that

are coming along.

So that's the kind of stuff that rattles

around in my head.

So

Yeah,

you started this conversation some weeks

back, Brian,

and I remember one of the ways you

phrased it is looking at drug development

as a math problem.

And so I thought, well,

maybe can I come up with something that's

more of looking at drug development as a

data problem and trying to figure out,

okay, what exactly do I mean?

Right.

And so I spent parts of my career

trying to pull magic signals out of omics

data,

trying to be as certain as we could

or bring as much data as we could.

But then it was interesting.

I, you know, I flipped,

started being involved more in the later

stages and thinking, you know,

there it was less, you know,

hypothesis generation from omics data sets

to kind of probabilities of success.

And you started,

so I got exposed to this style of

thinking realized how big the problem was.

And, you know, one of the things,

you know,

I'm just looking at some notes here then,

right, that, you know,

how this number of did it gets thrown

around at ninety percent of clinical

programs fail and

That all,

it depends on how you do the accounting,

but there are ways to do the accounting

where that number is very real, you know,

yeah.

Eighty to nineteen.

Right.

Yeah.

Um, what were, so first off, you know,

forty half of its laugh of efficacy,

a third was about safety toxicity,

and then you had, you know, the,

the rest maybe operational strategic

commercial.

Um,

and then I was digging in just a,

a little bit more and.

trying to look at some of these later

stages where a lot of the money is

being spent.

And it was like something like over,

you know,

twenty five percent of phase three's

failed due to commercial reasons.

And like, well,

that doesn't have a lot necessarily to do

with some of what you were trying to

dig in at.

Right.

So let's keep that in mind.

you know,

fifty to twenty percent still due to

safety issues, but, you know,

arising kind of as you go to a

larger population.

So how generalizable and how well do you

understand that efficacy still remained on

the top spots?

But I think

one of my experiences then at the time

was we we ran a data science challenge

i collaborated with mit professor low

there who was in a healthcare finance

curriculum um and we were modeling the

trial probability of success and looking

at what was most predictive for you know

of whether trial succeeds or not and of

course like having a biomarker was

important it kind of doubled your

or brought you from ten percent success to

twenty percent success baseline chance by

just having a biomarker.

And I think that's important.

Because in my mind, it's not just like,

oh,

we got to go out and find a

biomarker.

I mean, if you can, great.

But if you're doing your drug discovery in

such a way that a biomarker is coming

along,

inherent to the way that you're doing

things, because you're looking at,

you know,

patient populations along with omics,

and you know, you,

your identification of that target was,

you know,

as a result of some prediction predictive

features that were becoming that biomarker

so I think maybe that's where I think

it is is the mostly useful to think

about and then when we did this

probability success challenge it was

really about like clinicaltrials.gov you

know structured data and that's okay but

if you want to start thinking about this

you know,

maybe more in a Bayesian way or,

you know,

simply in an integrated evidence way,

could you start to extend that?

And that's one of the things we were

hoping to come out of that.

I mean, as you were talking about,

you know,

drug development discovery by nature is

heavily siloed.

You know,

one team has one accountability

responsibility, do the best you can,

throw that out the other side.

And now another team picks it up and

tries, works on it there,

putting out something high quality at the

other side.

But you could extend this.

What is the probability that we have a

translatable successful registration and

bring this all the way back to your

target ID in some senses?

And maybe that's how an AI would tend

to look at this problem.

And so the Bayesian way would be a

little bit

more along the lines of this.

If this is a good target,

what would we expect to see?

What would we need to see?

What would be the risks and what

observations would let us think that,

you know,

these risks have been mitigated already?

Right.

So it doesn't have to be Bayesian,

but just, you know, think of genomics.

It's common these days, you know,

especially with all of the populational,

you know,

Population level kind of genomics

databases out there, you know,

to try and triangulate and see some

genomics evidence, you know,

to so you can see function rather than

just sort of correlations or function in a

real world population.

Through an alien randomization, or,

you know,

whatever sort of signal you want to think

about, you know, and, you know,

just thinking from the.

clinical development teams modeling

probability of success for decision

making.

Can you start going backwards,

taking into account in a math problem like

ways, some of the preclinical data,

the admit data, the sensitivity data,

just propagating this farther and farther

backwards.

And if you could do this,

it could probably help you optimize what

data you were generating,

how you were spending your time,

but also help you optimize all of the

decisions along the way.

I think, you know, that's what your,

you know, thoughts, you know,

that seed triggered in my head as well.

And I mean, for me, you know,

there's a couple of things,

things that came up, right?

Is that, you know,

even when you took out the percentages of

failure, right?

Is it efficacy, safety, right?

And obviously we need to put them in

those buckets.

But I think we still continue to do

that work preclinically in those buckets,

right?

We talk about the siloing,

but it's still so siloed, right?

So at the beginning, the decision,

when you get closer to your packages

submissions,

the decision is really not looking at is

there an effect,

but looking really at what exposure and

after what duration the system stops

coping.

And then so I think if we kept

integrating,

like thinking about digital measures,

right?

And Brian,

we spoke about that a little bit earlier.

You mentioned that integrating some of the

behavioral aspects and physiological

aspects to it, right?

So you can actually do both at the

same time.

And then so there is a lot of

benefit to that, right?

You can use less animals,

you can make decisions faster.

And then so to your point, then, Nick,

that you mentioned about the biomarkers,

right, that leads to the higher success.

And but very often preclinically, right,

we use different biomarkers and endpoints,

right, that we do in a clinic.

And that disconnect is huge.

And so there's that opportunity, right?

I mean,

if there's such a huge success rate,

we should focus more on a preclinical side

to identify biomarkers that we can track

also in a clinic.

And I think one area is, again,

the digital measures, right?

I mean,

we have so many different endpoints now

that we use in a clinic,

but we still refuse to use them

pre-clinically.

Even the technologies here,

technology is being utilized.

And even for like safety,

I get pushed back all the time when

you talk about the non-invasive digital

measures.

uh because we have telemetry right and

then we you know come in and monitor

the dog every hour right to see if

they're vomiting you know how they're

behaving etc and we have technology right

for even safety assessment that you can do

that right the behavior and physiological

um which would give us this continuous

what we do in the patient so um

again it kind of just triggered those

those those two things to me um

Yeah, and Brian, jump in when you want.

There's a couple of thoughts before I lose

them.

One, we're getting at a couple of things.

One and two things.

One, translatability,

and I'll come back to that.

The first was that there's an interlinking

of efficacy and safety, right?

Because even though we tend to talk about

them as two different things,

and I think we might have mentioned this

before, they're one and the same, right?

It's all about dosing.

And I think there's a connection to

something that we were talking about in

one of the prior months.

Targeted therapies are one of the hot

topics right now,

whether they're bifunctional molecules or

lipid nanoparticles or very disease tissue

specific targets.

And so what that does is if you

can localize

you know,

all of your drug around a specific type

of tissue,

you need to need a much lower dose.

And that lower dose comes with, you know,

much less safety risk, right?

Or alternatively, you know,

you can get the optimal effect for a

given, you know, with a lower dose,

you know, lowering your risk of toxicity.

So that's, you know, not only one, but,

you know,

helps probably deliver those to the

exposure.

And that was the other one, right?

The risk is, you know,

safety is one risk.

The other thing is just the exposure to

the diseased tissue,

as we were talking about these types of

things, right?

So, Brian,

we were talking the other day about how

a lot of these things are measured

in blood and that's very well established

and you know the pkpd and these types

of things but you know it's a bit

more complicated once you start trying to

get into the the tissue or even the

exact cells or even the compartment in the

cell that it might need might need to

get to so what the what do data

sets look like you know that can help

us better understand and predict you know

what uh you know what can help boost

this exposure and then you know as you

guys were talking about um

not only is this target going to translate

into humans,

but also how many humans and how much

variability is there.

And even if it does,

is it going to solve the unmet medical

need versus just perturb some blood values

and what actually matters at the end?

Back to your comments about digital

measures and these types of things as

well.

You know,

you guys have introduced so many

interesting concepts and threads.

We could be here for easily the rest

of the day because I got all these

ideas rattling around in my head.

But there's two that I want to follow

up on because I think they're sort of

fundamental.

One is, as you both talked about,

not only

the how we structure data,

what the kind of data we're actually

collecting,

and then ultimately how you apply it in

both a prospective and I don't think we

do enough of retrospective application

actually learning from our experiences

because we're just so intent on,

turning the crank and getting the next

lucky molecule out.

I think that's something that pharma needs

to take more responsibility for,

which is to find ways of integrating data

and learning from it much better.

That gets into the siloing that Nick

mentioned.

a something that that's actually outside

of pharma that relates to our our approach

to health care.

And I tell you,

I don't mean it's just a United States

thing.

I think it's a global thing that actually

puts pharma at a huge disadvantage.

And again, in sort of trying to

work with my mom here,

I'm seeing it in spades.

And that is at the time point at

which we engage disease.

Because the way our healthcare system is

designed, our horses, our disease horses,

are way outside the barn before we start

to engage them.

And so you've eliminated a lot of your

opportunity to actually mitigate the

progression of that disease process in a

lot of ways.

Because a lot of what we're trying to

do is manage chronic progressive diseases

that are already very,

very well developed.

So I'm saying that to say that even

though it's entertaining for me to beat up

on pharma because I've seen firsthand the

things that I don't think we do very

well,

to some degree pharma has taken on a

challenge with debt pretty much stacked

against it.

And so I'm actually amazed that we do

as well as we do in the context

in which we're trying to do it.

And again, thinking about,

given our interests,

How you maximally leverage technology to

be able to address some of these

fundamental challenges, I think,

is the thing that we need to think

more about.

And some part of that is, again,

circling back to what you said in the

front end, Nick,

is breaking down some of the silos.

And so it's.

We live in a world where we've been

able to dive into incredibly deep

expertise,

but the disadvantage to that is that we've

got a lot of really,

really deep expertise that doesn't

integrate very well across those

disciplines and that expertise.

And we need to figure out ways of

dealing with that.

And so that makes me kind of think

about, again,

just trigger over the last couple of

years, right?

The two reports that I was part of,

from NIH and the National Academy of

Sciences,

one on non-human primates used in

biomedical research and the other one on

alternatives.

And so in both of those reports, right,

we're talking about combinatorial

technologies, right?

That, you know,

if you combine in silico and in vitro

and in silico and in vitro and in

vivo technologies like digital measures,

you know,

microphysiological systems and in silico

predictions, right?

You're going to have a bigger impact,

right?

And better understanding

of what's happening versus individually,

right?

So they're silent.

And I think the same thing here, right,

is taking this combinatorial approach,

right,

as we're looking at combinatorial from the

perspective, hey,

we can monitor for safety and efficacy.

And I know there's areas where, you know,

we have tried to do that, you know,

for decades, I guess now,

different areas that come in,

especially from the safety side.

But it's not a typical thing that we

do, right?

And this, again,

the combinatorial of using digital

measures from the clinical side, right,

reversed into the preclinical side.

And then so,

I'm sorry because I'm getting distracted

by Nick typing, but...

I was going to lose my thought otherwise.

I had to get it down.

That's why God made pencils.

Yeah,

I think I'm going to have to send

that to you, Nick.

But anyway, so I think, yeah, again,

these combinatorial and de-siloing aspects

that we already touched on.

And then that's going to, you know, Brian,

like you said, the math problem.

But then it also becomes more challenging,

right?

Because then you have all this data,

right?

So how do you analyze it?

What do you do with it?

Is it useful or not?

And it's a significant investment.

And very often, again,

when that comes around,

we end up kind of pushing back, well,

you know, there's so much, right?

So we're not sure what to do with

it.

And potentially could have a negative

impact on us to move something forward.

So let's just not do it, right?

I mean,

microphysiological systems were a really

good example of that.

Yeah,

we leave a lot of value on the

table because we don't have good ways of

being able to mine it from all those

experiences.

I'm going to rewind a little bit to

something Brian was saying.

He was almost alluding towards kind of

you know, prevention,

his disease horses being outside the barn,

right?

So I'm going to go back to the

disease horses here.

And there were two examples that came to

mind.

And I think both of them can be

funneled back, as you say,

in this kind of learning culture back to,

you know, how you do drug discovery,

you know,

ideally, you know,

the ideal of pharma therapy is something

that can halt progression or even reverse

it, not just, you know,

but sometimes it's non-reversible.

You can only halt it,

but not just treat the symptoms, right?

So like, okay,

you at least halt progression.

But we had one and I won't mention

the indication, but, you know,

the target ID was done by looking at

kind of late stage patients when the

damage, phenotypic damage had been done.

And it was, you know,

the team felt it was like a real

preventative therapy but in order again to

de-risk the trials you know you would need

you know there's not going to be much

of effect if there's been a huge amount

of damage done but if you could catch

somebody before that you know while the

horses are in the barn there's going to

be a bigger signal that you could get

except the problem was you know like there

was yes a late stage disease label and

then there was like everybody else so like

we almost and when i say we needed

to invent an intermediate like it's

you know,

it's not really inventing a new clinical

parameter,

but you had to understand and be able

to predict who was at risk of being

a progresser.

And so, you know, that was the task.

And, you know, imaging was, you know,

asked of kind of the data science and

A.I.

team in this particular case.

And so that was a fun project.

But, you know,

I think even from being able to do

that one is like, well,

you enable, you know,

you de-risk the trial,

but you also make that therapy a lot

more valuable.

But I think there's learnings you can

probably take back from that too.

But even just being able to diagnose

people earlier or something.

The other example I wanted to say is,

you know,

talking to a friend at a health system,

you know, or just clinicians,

you realize like to take an indication

like depression,

once you become clinical and gone in to

seek help,

I am told,

it seems to be accepted that it's much

harder to climb out of that depressive

hole than if you could have been prevented

from going into it.

And so the thought was,

can we use some digital measures actually

to help flag or interact or monitor

patients?

You might be at risk.

But then my thought there was,

if you can create that data set and

you know this progression patients go

through,

that can be an effective diagnostic

screening tool.

then you could look at what types of

patients and, you know,

can that be fed back into drug discovery

or at least in a clinical design program

as well,

if you've actually captured that data.

You know, clinical management of pain,

I think,

is one of those areas where we've done

a really good job in being much more

proactive.

And you hear that a lot, particularly,

you know,

if you're a patient in the hospital,

even engaging a PCB is that you need

to engage pain early and often.

and not let it get out of control

because it's a whole lot harder to

backtrack it.

When I was in my early career,

I used to study chronic progressive renal

disease,

and we had this concept called final

common pathways.

And so you mentioned fibrosis.

Fibrosis is a common final pathway for a

lot of chronic progressive parenchymal

kinds of diseases.

And a lot of our drug development focus,

particularly for fibrotic diseases like

IPF and chronic renal disease,

has been on those kinds of mediators.

Well,

you're never going to turn those back.

That is the common final pathway.

You really need to understand the front

end.

But the rub of that is,

is that we don't oftentimes understand

some of those primary etiologic events

that actually drive you to that place.

And so I'm saying that to say that

we need to put more emphasis in studying

clinical disease.

But that's only helpful if you can detect

patients early in that clinical disease.

So you get a better understanding of the

early stages so we can start.

engaging it earlier.

And I guess, so for me,

that kind of triggered, Brian, I mean,

there's a right, because you said,

you know, managing that pain, right?

They were doing, you know, starting to do,

you know, a better job at it, right?

And just the only thing that flashed in

front of me.

Right.

It's the smiling and a frown face.

Right.

Do you go to the hospital?

Right.

And hey, tell me from zero to ten.

I mean,

it drove my mom crazy when she was

undergoing chemotherapy.

Right.

I mean, it was ridiculous.

And and well, it's even, you know, really,

to be honest, just pissed me off is,

you know,

we were at this meeting yesterday.

I won't say which one, right?

But so we're talking about digital

measured technologies, right?

And we're trying to, you know,

talking about for activity level,

for the sleep, right?

So I asked like, you know,

why is nobody looking for assessing pain,

right?

And the response from everybody was it's

too difficult, right?

And, but I mean,

it's such a huge impact, right?

I mean, pretty much almost every disease,

right, that you have.

I mean, we're looking at fatigue, right?

But a lot of times that's because of

pain, right?

I mean,

pain has impact on so many different

things.

And it was pretty much, you know,

it wasn't like even people thought about

like, oh,

maybe we should think about doing that.

It's just it's just too challenging.

We're not even touching it.

And so that's kind of really,

really frustrating.

And that even then also switches to me.

Like when we talk about behavioral digital

measures that we collect now, again,

even non-invasive one, right,

for behavioral studies.

But when you try to talk to people

like in a cardiac, right,

if you're doing a cardiac surgery,

You should assess those, right?

Because you have behavior changes,

fatigue.

I mean,

that's what we're doing to patients now,

right?

It's like, well, no, we're cardiac, right?

We need telemetry and that's it, right?

I mean,

I just still don't grasp the concept,

right?

We're trying to, you know,

and regulators are pushing, right?

And even insurance, the payers, right?

We want a holistic picture of the patient,

right?

and pre-clinically we're like no we're

doing cardiac so let's do cardiac we're

doing respiratory let's do respiratory um

and even there sorry that another trigger

right in respiratory studies we're

monitoring body weight i mean what that

that really um where we can monitor

respiratory rate you know continuously um

so yeah it just had a couple triggers

there you know it's a good point because

we settle on biomarkers that are easily

accessible rather than

um particularly relevant right to the

point you just made you know and then

that preclinical to clinical um

translation you know we have the benefit

of doing a lot of morphologic you know

uh

post-necropsy kind of assessments.

And I tell students this all the time.

I said, you know,

histopathology is a biomarker strategy,

but patients don't like it very much when

you start cutting out pieces of their

organs.

That's not a good clinical biomarker

strategy.

So you need to come up with things

that are much more clinically

translational.

It's high resolution.

It's very informative,

but it's not very translational.

Now, and so the other one also, Natruga,

is right.

So preclinically, right,

the way we assess, you know,

it's cortisol.

I mean,

you have to pick up the animal if

it's in pain.

You're grabbing the animal, right,

collecting the blood, right,

which is an invasive procedure, right?

And then you're going to tell me where

the animal is, I mean, like, you know,

stressed or, you know, and et cetera,

right?

we don't use it because it's all over

the map because for the reasons you just

mentioned yeah yeah there's a ton of

papers right utilizing that right as an

assessment um of it of stress right and

and and discomfort yeah anyway it's yeah

chime in with something here too and maybe

it's a good food for thought for one

of the upcoming uh months um you know

you talked about you know why don't people

factor pain into account and you know your

mom there and it made me thinking i've

been you know i talk a lot about

precision medicine lately less about

personalized from a pharma setting maybe a

lot about personalized from a digital

health setting but you know i was trying

to think in that case what would a

personalized

because we were talking about what does

that even mean uh solution mean even

earlier what does personalized medicine

even mean anymore um you know versus

precise you know would it have been

something on the molecule maybe not you

know would it have been something in the

dosing of the dosing guidelines or does it

start to be more of a feedback and

kind of a decision tree on you know

how to get the dose right for somebody

to mitigate the pain of what they can

talk like all of these types of things

it's like well

The care, the administration, like the,

you know, from adherence to dosing to,

you know,

around the pill support coaching,

I don't know,

but these are all things that can be

easily,

the implementation of that drug can be

easily personalized, I think,

in today's world with all the data

feedback loops that we have.

And, you know, on the molecule side,

I don't know.

But it's like, well,

what does personalized mean in molecules?

Maybe it's less about reacting to an

individual,

but understanding the genetic makeup and

lifestyle.

Like, I don't know, right?

There's certainly,

you can have personalized treatment

decisions.

Is personalized in molecules just having

so many different

varieties that you know like you said in

fibrosis you can you know distill down and

get beneath the phenotype to more root

moas and maybe some that even you know

oncology is all kinds of you know hair

too and you know biomarkers that you can

get enough therapies to choose from to get

the one that's best i mean if we're

to do that rest of us probably have

to improve this ninety percent number we

said and get like you know a lot

more things faster more efficiently

through that pipeline um and

you know,

commercial viability matters too, right?

So, you know,

or is there more of a process?

It would be great to have one,

you know, think of mRNA vaccine style,

or CAR T, you know, and like,

things which by default might be

personalized to you if they're, you know,

if it's coming a little bit, you know,

it's a process that takes

you know,

some bit of tissue or a cell from

you and trains an immune attack to say

like,

then it is like hyper personalized by

default, but like, you know,

can you cascade back from there a little

bit?

in terms of processes that are easy to

either train, expand cells,

print slightly different sequences,

optimize.

I don't know.

We've talked about personalized medicine

for decades.

What do we even mean anymore?

And if we step back and think big

picture,

what is this actually you know maybe we

achieve personalized medicine and it looks

nothing like all of us had imagined in

our heads i guess that was what i

was starting to to think now with some

of your comments i don't know what your

your two thoughts on this are but

I mean, one of the things, right, that,

you know, it's the terminology also,

right?

We very often fight about that, right?

So you're talking personalized, right?

I heard, you know,

some speakers talk about, you know,

I never use personalized,

it's individualized, right?

Which again, like to me, it's like,

you know, again,

I don't care which one you use, right?

But let's talk about, you know,

how do we help patients, right?

And so now you're talking about NAMs,

right?

I mean, you know, it's the same thing,

right?

We spend all this time trying to define

NAMs and we're wasting time, right?

Instead of trying to help patients.

So to me that again, I mean,

I guess to today through our sessions,

right,

there's a lot of triggers that come in

that.

Yeah.

And so maybe even it's, you know,

there's two ways, you know,

one is

decisions or input strategies that involve

a person's genetic background and medical

history in terms of selecting a therapy.

And then another is our ways to adapt

based on feedback loops from digital

sensors, imaging,

other things in today's world that can

adapt a treatment course.

So maybe there's two different ways of

looking at it.

You know,

to sort of bring this back to the

original theme of drug development as a

math problem or thinking more

quantitatively,

I think a lot of the things that

we're talking about all relate to this

idea that we really do need to have

a more

cradle to grave, as it were,

thought process around how we're going to

engage disease,

and then ultimately try to figure out how

to use technology to create a better

integration that allows us to not only

prospectively make more

better, more predictive decisions,

but actually allows us to go back and

learn from those experiences as most of

them aren't successful.

And so you're completely losing the value

of that lack of success if you can't

go back and mine that better.

So I think that's a common theme in

all the kinds of things that we just

talked about.

It is basically trying to think about how

to make what we do more quantitative so

that the technologies we have available to

us have maximal impact.

And right now we have a system that

is probably still too siloed,

even though we've had that discussion for

the decades that I've been involved in the

business.

We've recognized it.

We've talked about it.

And we just struggle to get past it.

so that sounds like a good i don't

know how many minutes we're supposed to

take on our little chats here but that

sounds like a pretty good weaving together

of all the different kind of points we're

making so yeah i think that well i

hope one i hope that folks have found

this interesting as we've we'll do more

scoping of problems and solving them right

that's kind of what we do but we

look for that that interesting

intersection

But I expect that some of the threads

that we've raised today will end up being

the focus of some of the stuff we

write in our future essays.

And I'm looking forward.

I'm so jazzed.

I've got the paper and pencil right here.

At least it's good.

It's good that Nick finally stopped

typing.

Thank you.

Well, with that, I said, yeah,

we're going to say goodbye to you guys.

Well, to you, Nick and Brian,

and also to folks who are listening to

us and looking forward to seeing you guys

at our next episode of Innovation to

Impact.

Cheers.