Episode Transcript
Transcripts are displayed as originally observed. Some content, including advertisements may have changed.
Use Ctrl + F to search
0:15
Pushkin. Over
0:21
the past few decades, it's become more and
0:23
more expensive to develop new drugs.
0:25
It now costs over a billion dollars
0:28
on average to bring a new drug to market
0:30
in the United States, and of course
0:33
drug companies pass those high development
0:35
costs onto us in the form of higher
0:37
drug prices. This has been
0:39
going on for so long that we have sort of gotten
0:41
used to it. But when you zoom out,
0:44
it's strange because,
0:47
as I've said before on this show, and as
0:49
I will say again on this show, one
0:51
of the main things technology does
0:54
is it makes things more
0:56
efficient and therefore cheaper.
0:59
Over the past few centuries, we've seen technologies
1:01
make all kinds of things cheaper, everything from
1:04
clothes to food to TVs.
1:06
So why hasn't new technology
1:09
made drugs cheaper? Two. I'm
1:15
Jacob Goldstein and this is What's Your Problem,
1:17
the show where I talk to people who are
1:19
trying to make technological progress.
1:22
My guest today is Alice Zang, co founder
1:25
and CEO of verge Genomics.
1:27
Alice's problem is this, how
1:30
do you use artificial intelligence to
1:32
drive down the price of discovering
1:35
and developing new drugs? Why
1:38
is it getting more expensive to develop drugs,
1:40
despite the fact that we have better technology to do it.
1:42
Yeah. Absolutely. One of the reasons
1:45
is, you know, even though a lot of the new
1:47
technologies we've developed have made
1:49
us better at
1:51
testing more drugs faster, but
1:54
the fundamental problem is that even
1:56
if we can get a drug all the way to clinical
1:59
trials, which is the last step of drug development,
2:02
ninety percent of those drugs still
2:04
fail. So if you think about it, we're spending millions
2:06
on each drug. Of
2:09
those drugs are failing at the last and most
2:11
expensive stage of drug development. And
2:13
so really most of that billion
2:16
plus dollar figure you hear is
2:18
due to the cost of failure. Just
2:21
to be clear, that figure more than a billion dollars.
2:23
It's you've got to include the cost of
2:25
all the drugs that don't work exactly,
2:28
the ones that do right exactly. So the ones
2:30
that do work have to pay for all the ones that fail.
2:32
That's the fundamental problem, exactly, And
2:35
you're setting out to fix that if
2:37
you can. Absolutely, we think
2:40
there's an opportunity for
2:42
AI to fundamentally
2:44
shift really the failure
2:46
rate, and the most impactful time
2:49
to do that really is the failure in clinical trials.
2:51
So can we predict
2:53
before we go in to these
2:56
expensive clinical trials genes
2:58
or targets or drugs that are more likely
3:01
to work in humans, because even a
3:04
ten percent decrease in that failure
3:06
rate could have massive
3:09
I saw a number of up to fifteen
3:11
billion dollars annually in industry
3:13
cost savings. You could still
3:15
be in a universe where most of the drugs
3:18
that go into clinical trials fail, but instead
3:20
of ninety percent of them failing, seventy
3:22
percent of them fail, and that would be a huge win.
3:24
That would be a huge efficiency gain. It would save a
3:26
ton of money, absolutely, And I think that's
3:29
something that's underappreciated about
3:31
AI and really any technology, is
3:33
that oftentimes people have this expectation
3:36
that this technology is going to absolutely
3:38
transform a field overnight. And I
3:40
think what people don't appreciate is that most of
3:42
the time that doesn't happen. It's always step
3:45
by step incremental. But even a
3:47
ten percent change would have billions
3:50
of dollars of cost savings and
3:52
would be a huge win for patients
3:54
in the industry worldwide. I like that
3:56
frame, actually, I like that frame of maybe
3:59
AI can have drugs fail
4:02
most of the time, but not as much of the time
4:04
as they fail. Now, like, it seems very credible,
4:07
It seems very plausible. Would you put it that way? Yeah,
4:09
it's all life is nothing but a learning process,
4:13
Yes, getting less bad at everything.
4:16
So I know you were
4:19
studying to be a doctor and a researcher not
4:21
that long ago, a few years ago before you started your company.
4:24
Like, tell me how you went from an
4:27
mdphd program to starting the company.
4:30
Well, my PhD research was
4:33
actually in using genomic analysis
4:35
and computational biology to
4:38
analyze large scale data sets
4:40
and find new drugs that
4:43
could improve drug development.
4:45
And we found that from our
4:48
very first drug that was predicted
4:50
from our algorithms when we put it in mice
4:53
after they've been injured, help them walk
4:56
and recover from that injury, that
4:58
nerve injury about four times faster than the leading
5:00
standard. And that was just the first drug that was
5:03
predicted. And I looked at this technology
5:05
in this approach and I thought, Wow, there's so much
5:07
promise here. You know, am I really
5:09
going to just publish this and
5:12
let it sit on a bookshelf somewhere,
5:15
or if I'm not going to be the one to
5:17
really develop this to patients, you know
5:19
who will, And when I looked out
5:21
off the field, I did not see a
5:23
ton of biotech or farmer companies
5:26
that were truly computationally driven.
5:28
Usually within pharma companies they might bring
5:31
in computational biologists to support.
5:34
There are scientists or their biologists, but
5:36
there wasn't really a genomics
5:39
computationally driven company at that time. Now
5:41
there are many, but at the
5:43
time there are very few. And so I
5:45
actually, you know, it wasn't a binary decision.
5:47
People always ask me, how did you make the
5:50
courageous decision to leap?
5:52
It wasn't really like that.
5:55
I think what we did first is that we
5:57
just took three months three month leave of absence.
5:59
We joined a program, an incubator called a y
6:01
combinator. We as you and
6:04
you and well me and my co founder Jason.
6:08
And the first question really was, you
6:11
know, can we even generate
6:13
some data that validates that
6:15
computational biology can predict targets
6:18
that work? And then when we saw
6:20
some data, the next question was can
6:23
we even hire people that want
6:25
to come on? And the next question was can we
6:27
even raise money from people
6:29
that will care? And I think that
6:32
is so such an important lesson because
6:34
I think people oftentimes get caught up in
6:36
just the destination, you know, is where
6:38
I want to be? Is this the career I want to have
6:41
that they don't take the first step,
6:43
And really it's the first step that's needed to
6:45
actually get the data to even decide if
6:48
it's the appropriate track for you. And
6:50
did you really just keep thinking, well, this might
6:52
not work, but let's do the next thing. Were
6:54
you in a place where you could have gone back to the MD
6:56
PhD program for a while. Yeah.
6:59
I took a leave of a continuous leave of absence
7:02
for probably over five years,
7:04
probably more than I should have, until
7:07
the point where a lot of my friends are like, are you really,
7:09
are you really gonna go back? And
7:12
finally the medical school is like, you're not really
7:14
going to come back, let's just terminate
7:16
your leave of absence. But it was in
7:18
the first few years a really important safety net
7:20
for me that gave me the psychological
7:23
safety to really take a risk and
7:26
really pursue a new idea that I don't
7:28
know if I would have otherwise. And I think
7:30
that's so important. I think for universities
7:33
to provide is that to recognize there
7:35
can be more than one track for
7:37
people to do really excellent science
7:40
and make an impact more than just becoming a
7:42
professor. And sometimes that
7:44
psychological safety is what's needed
7:47
to help people find their ultimate
7:49
calling too. By the ways, so
7:53
far, By the way, what's
7:56
a very brief definition
7:58
of computational biology. It's
8:02
really, at the end of the day, in my view, just the
8:04
use of computers
8:06
and data sets to understand
8:09
and biology better. By the way, what
8:12
happened to that molecule that
8:15
you were testing in mice in grad school?
8:17
That seemed useful? I
8:20
don't know. It's a good question. Actually, I
8:26
think the project was taken on by someone
8:29
else, but I'm not actually completely
8:31
sure. So, Okay, you
8:33
leave grad school, you start a company
8:36
you in fact now have taken
8:39
You have a bunch of molecules that you're
8:41
working on, and that seemed promising. But there's one
8:44
that is in clinical trials now
8:47
right to treat als Luke
8:49
Gary's disease, a terrible disease that is very
8:51
poorly treated. And
8:53
I thought that we could talk about the
8:57
story of that molecule of that drug
8:59
as a way to understand the way your company works.
9:02
Can you just sort of take me through the life
9:05
of that drug? So far? Yeah,
9:07
absolutely. I'll start off just
9:10
by talking about als and
9:12
why it's been so hard to
9:15
discover the right therapy, and then you
9:17
know why how we did that differently. So,
9:19
as you might know, LS Luke
9:22
Garrig's disease is a really horrible disease.
9:25
What happens is that these neurons
9:27
called motor neurons start dying, and
9:29
most patients experience paralysis
9:31
and then death, usually within three to five years
9:34
of diagnosis. A very fast progressing
9:36
disease, and there really aren't
9:39
any meaningfully effective treatments
9:42
that really slow or stop the disease today.
9:44
So a very horrible disease with
9:46
a horrible prognosis and no available treatments,
9:49
and why it's been so hard I
9:52
think to discover really effective treatments
9:54
is really just the complexity of the disease,
9:57
and really any disease of the brain, the brain
10:00
is the most complex organ in the body.
10:02
So you end up having a lot of drugs
10:04
brought into clinical trials that worked in mice.
10:07
I always like to say we've cured LS or can There
10:09
are many diseases in mice a thousand times,
10:12
but none of them have really worked in humans.
10:15
So what we did differently was we started
10:17
from day one by collecting data
10:19
from over a thousand ALS patients
10:21
as well as controls, and specifically,
10:24
we collected samples of brain tissue
10:27
as well as spinal cords from these patients
10:29
that actually passed away from ALS.
10:32
So you got samples from
10:34
a thousand patients who
10:36
had died of ALS. How
10:38
did you do that? So what we've
10:41
done over the last seven years is we've signed
10:43
partnerships with over twenty
10:45
one different brain banks, hospitals,
10:48
labs, academic centers worldwide
10:51
that collect these brain tissues. They're usually
10:53
donated from patients that have passed
10:56
away from the disease and whose families want
10:58
to contribute to research. Could
11:00
So step one basically is
11:02
get tissue samples from real
11:05
patients. And you said controls
11:07
as well, right, So tissue samples from healthy people
11:09
as well, so that you can use them as a basis
11:11
of comparison. You have the samples,
11:13
Now, what's step two? So step
11:16
two is that we put an
11:18
enormous amount of effort into quality controlling
11:21
these, So that's a big
11:23
underappreciated step. They can be very noisy
11:26
samples. And then step three
11:28
is that we sequence them, so
11:31
we profile,
11:33
what is the expression of all twenty thousand
11:35
genes in the genome, and we also
11:38
sometimes do DNA sequencing,
11:40
we look at genetic mutations. We
11:42
also have a clinical information
11:44
about that patient, how long did they have
11:47
the disease, when did they die? And
11:51
that makes for a very rich, multidimensional
11:53
data set, and that gives us essentially
11:55
a global snapshot of what happened
11:57
in that patient. H okay,
12:00
and you and presumably the
12:03
sequencing that you're doing on
12:05
the patient's tissue samples, you're doing the same
12:07
sequencing on the controls, the samples
12:09
from healthy people. So
12:11
now you have this very large
12:14
data set. What's the next
12:16
step. So then you have this snapshot
12:18
of what happened, and the tricky part is
12:20
to figure out what caused it. I often liken
12:23
it to a plane has crashed, right,
12:25
You're looking through the rubble and you want to
12:27
figure out how the plane crashed and how
12:30
that information can be used to prevent further
12:32
planes from crashing. So that's
12:34
when our software engineers
12:37
and data scientists as well as machine
12:39
learning scientists come in and we have
12:41
algorithms essentially to integrate multiple
12:43
data types all the way from the
12:46
RNA, so how the genes were expressed to genetic
12:48
mutations to essentially
12:50
create a map of disease
12:52
biology, and within the map
12:54
our networks of genes that are all
12:57
interconnected that we believe
12:59
cause disease. And so I like to
13:01
think about it like when you're looking through
13:03
a plane crash the rubble, you want to find the black box,
13:06
which I'll help you figure out the cause
13:09
of the disease. And by having all the information,
13:11
we essentially locate the black boxes
13:14
of disease, the targets that are really
13:16
at the center of those networks, and
13:18
then we design drugs against those targets
13:21
that we believe can reverse disease.
13:23
It seems like differentiating
13:25
between correlation and causality in this
13:28
particular setting would be really hard, right, Like to
13:30
use the plane metaphor, if you had a bunch of planes that
13:32
crash and a bunch that hadn't crashed, you might say, oh,
13:34
like the wings were off all the ones
13:36
that crashed, and that's why they crashed.
13:39
But actually the wings came off because they crashed,
13:41
right, and it was something else that caused the crash. I feel like
13:43
that would be I mean, an
13:45
obvious problem. Yeah, that might
13:47
be hard. To solve Absolutely, you hit
13:50
the nail on the head, and actually the plane metaphor is
13:52
a really great one here. For one of the biggest
13:54
challenges with looking at tissue
13:56
from a patient that already died is that you're
13:58
getting the crash right. You're not seeing video
14:01
of before the crash. You're
14:03
really getting the crash. And the challenge
14:05
is how do you figure what caused the crash
14:08
versus as well was just the effect
14:11
of the crash, like a burned wing, etc.
14:15
And one of the ways we do
14:17
that is we combine different
14:19
data types. So we found that looking at
14:21
one type of data, for example,
14:23
just RNA data is in particularly
14:25
helpful, but it's actually looking
14:27
at where do you get convergence signal that
14:30
pulls through multiple types of data to
14:32
start revealing more compelling signal.
14:34
So as an example, we look at genetic data
14:37
as well. So genetic data is useful
14:39
for looking at cause versus effect
14:41
because it contains information about genetic
14:44
mutations that you were born with as a baby
14:47
that then lead to increased
14:49
risk later in life for a disease. And that's kind
14:51
of nature's human experiment
14:54
for really cause and effect. And
14:56
when we layer that on that information
14:58
on with the RNA data. It
15:00
actually gives us information about how the genetic
15:03
drivers are acting in these functional
15:05
pathways, which is a big issue
15:07
actually with just looking at genetic data on its
15:10
own. So I wish I had a better I wish I had a way
15:12
to actually string that through to the plane metaphor.
15:14
But and
15:17
there's a time for leaving metaphors behind. Your
15:20
company uses AI in drug
15:23
discovery. I appreciate in a certain way
15:25
that you haven't said AI yet, but
15:27
also I don't want to not talk about it. I
15:29
mean in the sort of figuring out what's
15:31
going on in this step? Is that well,
15:34
is that the first instance in this process
15:37
where you're using AI? Is it? We're talking about
15:39
that here? Yeah, I mean,
15:41
I think AI is a really broad term for
15:43
any kind of process
15:45
where the computer is learning from
15:47
something. So there are all sorts of applications
15:50
of AI in this entire process,
15:53
for example, how we're integrating the data
15:55
sets together, how we're inferring
15:58
what are the central nodes
16:01
or the key targets. I
16:04
would say the most classical
16:06
use of A on the way that most people think of it
16:08
is then once we have this network of
16:10
say one hundred genes, how do we actually
16:13
find what the cause is? How
16:15
do we find what is the hub or the right
16:18
target to hit to turn
16:20
off or on all hundred of those genes.
16:22
And that's where machine learning and AI comes
16:25
in handy.
16:29
In a minute, Alice explains how this actually
16:32
works in the case of the ALS
16:34
drug verges working on. Now now
16:44
back to the show. So okay,
16:46
Alice and her colleagues at Verge have collected
16:48
all these tissue samples from ALS patients.
16:51
They've used the samples to generate this huge
16:54
data set that shows genetic variation
16:56
and changes in how genes are expressed,
16:59
along with lots of clinical data about
17:01
the patients, and then they
17:03
build these basically these AI models
17:06
to try to figure out where in
17:08
this complicated biological
17:10
process that's happening in this disease, where
17:12
they should try to intervene
17:15
with a drug, Basically where they should try and
17:17
target a drug. I think of this
17:19
oftentimes, like if you think of a map
17:22
of all the airports in the US, you want
17:24
to figure out how
17:26
to go after the hubs like Chicago
17:29
or New York. You don't want to go an
17:31
airport in Kansas or I will wouldn't be very
17:33
effective at stopping airplane
17:36
travel in the country. So there's
17:38
a lot of different pieces of information
17:40
that we collect to then infer
17:43
what are the best genes that are not only central
17:45
within this network, but also there's
17:47
independent evidence of a disease
17:50
causal effect or a relationship to disease.
17:54
And so you do all
17:56
that in this instance, and
17:59
what do you figure out? So what the
18:02
algorithms spit out is essentially a ranked list
18:04
of targets, all right, So these are
18:06
ranked list of targets that are predicted if we could
18:08
dry them, would restore
18:11
that network back to levels of healthy
18:14
people and potentially
18:16
slow or stop the disease. And
18:19
then what we do is we take those targets and we start
18:21
testing them in the lab, all right, So
18:23
we actually what is kind of cool about the platform
18:25
is we get all these targets from human brain tissue
18:28
and we also can test them in human
18:30
brain cells in the lab. So you
18:32
get a list it's basically genes
18:34
to target. You either it says upregulate
18:37
or make this gene express more or make this gene
18:39
express less. Is that basically what the AI
18:41
is out putting exactly, Like,
18:44
so how long in the instance
18:46
of this ALS drug. How long was the list? More
18:49
or less, our initial set of targets
18:51
was twenty two high confidence
18:53
targets, and then
18:56
we actually then generated another chut
18:58
choosing updated data of about
19:00
thirty more targets as well. And
19:03
what was really striking when we tested
19:05
these targets is that when we tested
19:07
them in the lab, we found
19:10
that on average over
19:12
sixty percent of them, though more recently actually
19:14
around eighty percent of them actually validated
19:17
in the lab, so they actually protected ALS
19:20
patient cells from dying, which
19:22
is very high. So we're really
19:24
excited that we're actually seeing very
19:26
robust validation of the computational
19:28
predictions, at least in the lab. Okay,
19:31
so you have this list, you're
19:33
testing it, something like half
19:35
of them seem promising, you said, sixty
19:37
percent seem promising. What happens
19:40
next? Okay, So what happens
19:42
next is that we so we test them
19:44
in these human brain cells. We understand
19:46
the mechanism. One
19:48
of the really interesting findings from this ALS
19:51
program and specific is that when we looked
19:53
at the network that we found in these patient
19:55
spinal cords, we found a new
19:57
cause of disease that was previously unknown
20:00
so most of the hypotheses
20:03
in ALS, where many of them to date, have really
20:05
been focused around these protein
20:07
aggregates, these clumps of proteins that we
20:09
can easily observe by ie that you see
20:11
in ALS patients. Right, A lot of them are
20:13
observational hypotheses.
20:16
But what we found by looking at a deeper
20:18
cut of the data is actually,
20:20
at baseline, most of these
20:22
patients actually had a baseline
20:24
dysfunction in their life csomal
20:26
pathway, which I like to call the garbage
20:29
disposal pathway. It's what is critical
20:31
to clear out junk from the
20:33
cell. And because patients
20:35
were at baseline vulnerable to
20:38
these toxic insults, it wasn't so much the protein
20:40
clumps that were directly causing it. It was because
20:42
they're already vulnerable to these
20:44
clumps of proteins that their
20:47
cells started dying. And is
20:49
the idea that the gene
20:51
you're targeting is causing
20:54
the cell's garbage disposal
20:56
to not work, right, Like you're trying to fix the garbage disposal
20:58
by targeting this particular gene.
21:02
Yeah, it's a central regulator of that pathway.
21:04
And it was also a target that was ranked I think
21:06
it was ranked number one or number two on the list. So
21:10
just to be clear, how how
21:12
do you get from
21:14
you know, so you have fifty or
21:16
so things to test, fifty
21:19
or so targets, something
21:21
like thirty
21:23
of them seem promising. How
21:26
do you decide which of those thirty to proceed
21:28
with? Yeah,
21:30
so that's a great question. We get asked
21:32
that a lot. I think at that point it's
21:34
a strategic decision. Right, you were a startup,
21:38
Right, we have to be able to develop things quickly
21:40
and capital efficiently. So
21:42
we were lucky in that sense that one
21:45
of the top targets was also a target
21:47
that already had where
21:50
the path to developing a drug was relatively
21:52
smooth, A lot was known
21:54
about that target. We could start doing chemistry
21:56
and designing molecules relatively easily,
21:59
and the target itself had actually been tested
22:02
in the clinic for other diseases, not
22:04
als, but things
22:07
like Crohn's disease and surrounds, so
22:10
we did know there was some safety data
22:12
around hitting that target. We
22:15
do then for targets where we
22:18
can't develop all of the targets, right, we
22:20
can only take focused bets for targets
22:22
where there's a bit more technical risk,
22:24
Right, It might be a bit more exotic. People
22:27
don't really understand how it works. There's
22:30
not a lot of tools out
22:32
there to really develop drugs
22:34
against it. That's where we might partner
22:37
with a pharma company to
22:40
develop those targets. And we have such
22:42
a collaboration with Eli Lily where we
22:44
developed our als target, but actually Lily
22:46
has the opportunity to essentially take
22:49
you targets number three through twenty
22:51
two plus and
22:53
choose four of them to develop themselves. Oh
22:56
interesting. So in that way, you're essentially
22:58
laying off the risk to this giant
23:00
pharma company that can afford to make more bets.
23:04
I'd say we're distributing the risk and we're
23:07
allowing us to really capitalize on the entire
23:10
opportunity all of the targets, because
23:12
it's impossible for any small startup to do,
23:14
you know, thirty different programs.
23:17
And it's actually in line with what a lot of pharma companies
23:19
are looking for. A lot of pharma companies are looking
23:21
for. What is that novel
23:24
target that no one else is working on that's
23:26
kind of unexpected, Where if
23:28
we could really get a competitive edge in here,
23:30
this would be really meaningful for a position
23:33
within within drug development in the next
23:35
ten years. Well,
23:37
and I mean it also seems
23:40
compelling because even though this
23:43
seems like a more promising way to do drug
23:45
development, drug development is hard
23:47
enough that anyone candidate
23:50
drug is probably not going to work, right.
23:53
Yeah, An any biotechniqus to be able
23:55
to have a pipeline and the ability to withstand
23:58
I think some failures because
24:00
I think it's unrealistic to expect one hundred
24:02
percent of what you try will work. But that
24:05
doesn't reflect on the technology
24:07
itself, and that can be something
24:09
unfortunate in biotech, where you know,
24:11
if the first thing fails, everyone's
24:14
all can be. It can be tempted
24:17
to say, oh, the technology didn't work, but in
24:19
reality, you think about how many different
24:21
drugs that pharmac companies test all the time. Right,
24:24
So I think really promising technologies
24:26
need to be afforded that runway
24:28
and that ability to really take multiple shots
24:30
on goal before you can get the end to really see
24:32
if it's working. Right. Well, I mean, if nine of
24:36
traditionally developed drugs fail
24:38
once they get to clinical trials, you
24:41
could be way better but still likely
24:43
to fail on anyone drug. Yeah,
24:46
Yeah, even a fifty percent would be huge, right,
24:48
but still that means one out of two drugs
24:50
will fail. Relative
24:55
to the world we live in now, a world
24:57
where one out of two drugs fail
25:00
could be a world where we get more
25:02
new drugs for less money.
25:05
In a minute, the Lightning Round including
25:08
the worst thing out being named to the
25:10
Forbes thirty Under thirty, and
25:12
the best thing about accepting that your
25:14
company might sail. That's
25:22
the end of the ads. Now we're going back to
25:24
the show. Let's let's
25:26
close with the Lightning Round. You
25:29
personally interviewed over a thousand
25:31
people when you were starting your company, as
25:33
I understand it, which seems very
25:36
intense. And I'm sure as if there's
25:38
anything in your life outside of work where
25:40
you've been that intense. Oh,
25:42
everything that is a core
25:45
to my being. If
25:47
you ask my spouse, you would say any new
25:50
game that we start playing. And I'm very competitive
25:53
and it's just part of my being. I iterate,
25:55
I get a lot of reps in He always
25:57
likes to make fun of me that I have an
25:59
AI in my head. I'm constantly
26:02
learning and improving the model until
26:04
eventually I become a lean
26:07
mean. We've been saying a lot of
26:09
Katan recently, and I think
26:11
if we him fifteen times in a row, So
26:14
yeah, I am very intense
26:16
and thorough in my life. Is
26:20
chat GPT overrated or underrated?
26:25
Both? Actually? I think it's both over and underrated.
26:28
It's overrated for some applications
26:30
and underrated for others. I
26:32
think it's overrated for things where there aren't
26:36
a lot of information available already
26:38
on that thing. I think
26:40
it's underrated for applications
26:42
at coding, where there's already a large body of
26:44
literature out there. So it's really good at replicating
26:47
things that exist, less good at discovering
26:49
new things that don't exist. I
26:53
read an interview where you said one
26:55
of the things you've learned as in
26:57
running your company is you learn to be okay
27:00
with your company dying with your company not
27:02
making it, which I found like very
27:04
surprising and interesting. Can you just tell me a
27:06
little bit about that. Yeah,
27:08
I mean, I think it gets to really the core
27:11
of how we drive our culture, which is I
27:13
think that soul for so
27:15
long companies have been driven through fear
27:18
and bravado of you know, we're crushing it,
27:20
We're pounding on our talking about how we're
27:22
crushing it, and less about emotional vulnerably and
27:24
introspection and self awareness, and
27:27
ultimately I found the thing that really
27:29
transformed my leadership style was
27:32
learning what I had grips over of
27:34
where I was really attached to outcomes,
27:36
And ultimately, I think for all CEOs, a
27:39
lot of that is tying meaning to
27:41
what happens with the company. If the company fails,
27:43
this means something about me as a person,
27:46
and I think that stifles a ton of
27:49
innovation and curiosity and tends to drive
27:51
those cultures of fear. So ultimately,
27:53
the thing, for example, that got me to stop micromanaging
27:57
was really being okay with the company dying, because
27:59
ultimately, what is micromanaging if not
28:01
just fear right or fear or control.
28:04
And once you let go of that fear and you recognize
28:06
you're just open to learning. You can still really
28:08
want the company to succeed, and you can be passionate
28:10
about it, but you're no longer
28:13
thinking, oh, I'm screwed, or like I'm
28:16
a failure if this fails, and that just opens
28:18
a whole new level of levity and lightness.
28:21
Nice. What's the worst
28:23
thing about being named to the Forbes thirty
28:26
Under thirty list? I
28:29
think they did a photo shoot where
28:31
there was a there was a
28:34
very revealing split on the dress, and I still
28:36
get constantly made fun of by my close friends
28:38
for that. What's
28:42
one example of a thing that went
28:45
wrong as you were building the company?
28:47
Something bad that happened? Oh
28:49
so many things. We had a whole period where there
28:51
was a ton of attrition and
28:53
people leaving, and you know, the first time
28:55
that happens to a founder can
28:58
I took it personally, It's like someone leaving your baby,
29:00
and you wonder why. That
29:03
was actually a huge growth moment for me
29:05
because I was for
29:07
so long trying to put for the strong
29:10
face. If it's okay, it's okay. And finally,
29:12
at the end of like a month of this, I just
29:14
sat in front of the company at an all hands
29:16
and I honestly I just broke down in tears. I said,
29:19
I feel like I failed you guys. You
29:22
know I'm still grieving this. I really don't
29:24
know what to do. And it was paradoxically
29:27
in that moment, most of the team really
29:29
rose up to the occasion and I found support
29:31
in ways I didn't even know where possible from the team.
29:39
Alice saying, is the CEO and co founder
29:42
of verge Genomics. Today's
29:45
show was produced by Edith Russolo.
29:47
It was edited by Sarah Nix and Lydia Geancott
29:50
and engineered by Amanda ka Wong.
29:53
You're always looking for more
29:56
guests for the show. If there's someone out there working
29:58
on an interesting technical problem with big
30:00
stakes, tell us about that person.
30:02
You can email us at problem
30:06
at Pushkin dot fm, or
30:08
you can find me on Twitter at Jacob Goldstein.
30:11
I'm Jacob Goldstein and we'll be back next week
30:13
with another episode of What's Your Problem.
Podchaser is the ultimate destination for podcast data, search, and discovery. Learn More