Latest / Core Memory / We Shall Finally Map The Brain - EP 36 Andrew Payne

Transcript

We Shall Finally Map The Brain - EP 36 Andrew Payne · Core Memory

0:00
SPEAKER_00: embed the brain in a swellable polymer,
0:02
SPEAKER_00: and then you add water,
0:05
SPEAKER_00: it swells,
0:06
SPEAKER_00: it makes the brain physically bigger.
0:07
You can make it,
0:08
you know,
0:08
these days you can make it 20 times bigger in every dimension.
0:12
SPEAKER_00: And that physically moves apart the neurons.
0:16
And so now two wires that were overlapping that you couldn't
0:20
resolve in a light
0:21
SPEAKER_00: microscope now can be resolved.
0:22
SPEAKER_01: So you're literally taking one of these samples and stretching it out.
0:27
SPEAKER_01: And I was just with Ed Boyden
0:30
SPEAKER_01: I didn't know this because immediately I was like,
0:33
SPEAKER_01: oh,
0:33
SPEAKER_01: well,
0:33
SPEAKER_01: you're going to get some crazy distortion when you start doing that.
0:36
SPEAKER_01: But I mean,
0:37
SPEAKER_01: he was telling me it's like still,
0:38
SPEAKER_01: I mean,
0:39
SPEAKER_01: they're finding it's like 99% holding scale and everything.
0:44
SPEAKER_00: It preserves the structure remarkably well.
0:47
SPEAKER_00: Even version 1.0 preserves the structure remarkably well.
1:07
Welcome to the core memory podcast,
1:09
which I never usually say,
1:10
but here we are.
1:11
SPEAKER_01: We're with Andrew Payne.
1:14
SPEAKER_01: You are the co-founder of 11 bio.
1:17
SPEAKER_01: Yes.
1:17
SPEAKER_01: Yes.
1:18
SPEAKER_01: And, and I wrote a story about you guys 18 months ago.
1:23
SPEAKER_01: It was back in December.
1:24
SPEAKER_01: Okay.
1:24
SPEAKER_00: Okay.
1:25
SPEAKER_01: Okay.
1:25
SPEAKER_01: So it wasn't even that long ago.
1:26
SPEAKER_01: All right.
1:28
SPEAKER_01: Life is a blur.
1:29
SPEAKER_01: But yeah, I mean, thanks for having me here.
1:31
SPEAKER_01: Thank you so much for coming on.
1:33
SPEAKER_01: I mean,
1:33
SPEAKER_01: I'll do the whole intro on you,
1:37
SPEAKER_01: but you work on a subject that is near and dear to my heart,
1:41
SPEAKER_01: which is imaging the brain.
1:42
SPEAKER_01: That's what E11 is doing.
1:44
SPEAKER_01: Yeah, you used to cover that back in the day.
1:45
SPEAKER_01: I did.
1:46
SPEAKER_01: And now I think...
1:47
SPEAKER_01: I think we got the brainiest podcast going.
1:50
SPEAKER_01: I mean, we go deep.
1:54
SPEAKER_01: I learned as I was researching you,
1:57
SPEAKER_01: which I didn't know before,
1:57
SPEAKER_01: you went to the University of Toronto for undergrad.
2:00
SPEAKER_01: Are you Canadian?
2:00
SPEAKER_01: I am.
2:01
SPEAKER_01: You are?
2:01
SPEAKER_01: All right.
2:02
SPEAKER_01: I didn't know that.
2:02
SPEAKER_01: You disguised it.
2:04
SPEAKER_01: Yeah, I grew up in a suburb of Toronto, Mississauga.
2:07
SPEAKER_01: Okay, okay.
2:08
SPEAKER_01: And then you end up at MIT.
2:09
SPEAKER_01: Well, eventually, yeah.
2:10
SPEAKER_01: Yes, yes.
2:11
SPEAKER_01: Yeah, I mean, this is the quick bit.
2:13
SPEAKER_01: We're going to unpack all this as we go.
2:15
SPEAKER_01: But you end up at MIT.
2:19
SPEAKER_01: You studied under Ed Boyden.
2:21
SPEAKER_01: Yes.
2:21
SPEAKER_01: And I was just with Ed Boyden.
2:24
SPEAKER_01: Excellent.
2:24
SPEAKER_01: Yeah.
2:25
SPEAKER_01: And he, he, we will have an episode with him.
2:27
I mean,
2:27
um,
2:30
it's hard to be more famous as a brain scientist than Ed is,
2:34
I think.
2:35
SPEAKER_00: And it was, you know, really
2:38
SPEAKER_00: a wonderful adventure to be in his lab.
2:40
SPEAKER_00: So I would love to dig into that too.
2:42
I would like to talk about that because I feel like the most
2:44
interesting people I
2:45
SPEAKER_01: tend to meet come from George Church's lab or Ed's lab.
2:49
SPEAKER_01: And there's this, you know, like- And I guess I did do both.
2:51
SPEAKER_01: You did both.
2:52
SPEAKER_01: Okay.
2:52
SPEAKER_01: Well, there you go.
2:53
SPEAKER_01: So this does qualify.
2:54
SPEAKER_01: I feel like in the sports world, people do these like coaching trees.
2:58
where there's like some famous head coach and then it's
3:00
everyone who studied under them.
3:02
SPEAKER_01: I think somebody should do like the George Church,
3:05
SPEAKER_01: Ed Boyden,
3:06
SPEAKER_01: you know,
3:06
SPEAKER_01: professorship tree at some point.
3:09
SPEAKER_01: Um, but
3:11
SPEAKER_01: You're young and you're very distinguished.
3:13
SPEAKER_01: I mean,
3:14
SPEAKER_01: you like invented,
3:15
SPEAKER_01: you know,
3:16
SPEAKER_01: important things that led to other big things around,
3:20
SPEAKER_01: around how we might understand the brain and image it and,
3:25
SPEAKER_01: and,
3:25
SPEAKER_01: um,
3:25
SPEAKER_01: like just super accomplished.
3:27
I mean,
3:27
I've,
3:27
I knew you and I we've talked,
3:29
but as I was reading,
3:30
I got,
3:31
I got more impressed.
3:32
SPEAKER_00: Well, I mean, there's a lot to do.
3:34
SPEAKER_00: So, you know, can't go too slowly right now, actually.
3:37
SPEAKER_00: Okay.
3:38
SPEAKER_00: It's like a really exciting time for neurotech.
3:39
SPEAKER_01: Which is pretty much the heart of my first question.
3:43
But I'm going to set the stage for people who are not total
3:46
brain nerds a bit.
3:47
SPEAKER_01: I mean, in...
3:50
there there's been this thing called the connectome which is is
3:55
roughly i mean
3:56
vaguely equivalent to like the genome of the brain we've wanted
4:00
to to create this
4:01
picture of a brain down to the neuron level this very detailed
4:07
map um i mean i
4:10
SPEAKER_01: guess like
4:11
SPEAKER_01: conceptually it sounds easy enough,
4:12
SPEAKER_01: but it turns out to be quite,
4:14
SPEAKER_01: quite difficult to do because I think we,
4:17
SPEAKER_01: we created the first full brain of a worm,
4:20
SPEAKER_01: the C.
4:21
SPEAKER_01: elegans.
4:21
SPEAKER_01: I think it was like 20 years ago.
4:23
SPEAKER_01: Longer.
4:24
SPEAKER_01: It was in the seventies.
4:24
SPEAKER_01: In the seventies.
4:25
SPEAKER_01: Okay.
4:26
SPEAKER_01: So we did that and that's like an accomplishment,
4:28
but then we kind of went all the way until last year or this
4:32
year even and did the
4:34
SPEAKER_01: fly.
4:34
SPEAKER_01: There was a bit of a winter.
4:35
SPEAKER_00: Yes.
4:36
SPEAKER_00: It took some time for the technology to catch up.
4:38
SPEAKER_00: You know, I think
4:39
SPEAKER_00: you know,
4:40
SPEAKER_00: in many ways,
4:40
SPEAKER_00: like the worm connectome was this incredible milestone along the way,
4:44
SPEAKER_00: but also it was like a cautionary tale.
4:46
SPEAKER_00: It's like, do not do this at home, kids.
4:49
SPEAKER_00: Like you need to do a lot of work.
4:52
It was before we really had like good computer vision or like
4:55
good industrial microscopes.
4:56
SPEAKER_00: And so,
4:56
SPEAKER_00: you know,
4:57
SPEAKER_00: actually,
4:58
SPEAKER_00: you know,
4:58
SPEAKER_00: getting into it just for a minute,
5:00
SPEAKER_00: back in the day,
5:01
SPEAKER_00: people would like take pictures of the C.
5:03
SPEAKER_00: elegans slice by slice using these electron microscopes,
5:06
SPEAKER_00: take pictures of them.
5:08
line them up into like a 3D volume and then do a color by
5:11
numbers with colored pens
5:12
to trace the neurons through these individual transparencies
5:15
because we just didn't
5:15
have the AI tools and the equipment that we now have that
5:19
allows us to be done at
5:21
SPEAKER_00: scale industrially.
5:22
And that's sort of like where the analogy between like
5:26
connectomics and genomics come in.
5:28
SPEAKER_00: In both cases,
5:28
you sort of have like industrial scale microscopy that allows
5:33
you to get lots of
5:34
SPEAKER_00: data.
5:34
SPEAKER_00: And then you also have like a
5:36
bioinformatics or neuroinformatics backend that allows you to
5:39
actually cope with
5:40
SPEAKER_00: this big stream of data.
5:41
SPEAKER_01: Right, and the amount of data is astonishing.
5:45
SPEAKER_01: How does one get a slice of a...
5:48
SPEAKER_01: Before I do that,
5:49
SPEAKER_01: okay,
5:50
so we're going to get to this point where people are taking
5:53
these very thin slices
5:54
SPEAKER_01: of mice brains,
5:56
SPEAKER_01: of human brains,
5:58
like basically using a razor blade almost or something even
6:01
better than like a
6:02
SPEAKER_01: diamond-tipped...
6:04
SPEAKER_01: Most often been a diamond knife.
6:05
SPEAKER_01: Yeah, and so they create the slice and then...
6:08
SPEAKER_01: You have to stitch all the slices back to,
6:09
SPEAKER_01: you have to image them and then stitch all those back together.
6:11
But how does,
6:12
how do you even get a slice of a worm brain to do this in the
6:16
seventies?
6:16
SPEAKER_01: What does that look like?
6:18
SPEAKER_00: Oh, I mean, it's very tricky.
6:20
SPEAKER_00: I mean,
6:21
SPEAKER_00: you end up doing a bunch at once in a big block,
6:23
SPEAKER_00: but you know,
6:24
SPEAKER_00: you find the right specimen and then,
6:25
SPEAKER_00: and then you,
6:26
SPEAKER_00: uh,
6:27
SPEAKER_00: uh,
6:28
SPEAKER_00: you know,
6:28
SPEAKER_00: there's a lot of trial and error involved.
6:30
SPEAKER_01: All right, what do we do at Core Memory?
6:32
SPEAKER_01: We cover innovative,
6:33
SPEAKER_01: fast-moving,
6:35
SPEAKER_01: forward-thinking companies,
6:36
SPEAKER_01: which is why Core Memory is sponsored by Brex,
6:40
because Brex is the intelligent finance platform for many of
6:45
these companies,
6:46
SPEAKER_01: 30,000 companies from startups to the world's largest corporations.
6:51
rely on brex's technology for their finances they've got smart
6:55
corporate cards high
6:57
yield business banking and expense automation tools that are
7:02
fantastic i hate doing
7:03
my expenses and brex's ai's and software run right through
7:07
those expenses figure
7:09
out where we're spending money and take care of so much stuff
7:12
for you so you don't
7:13
SPEAKER_01: have to waste your time on it yourself
7:16
Go to brex.com slash core memory to learn more and just,
7:20
you know,
7:21
get with the program.
7:22
SPEAKER_01: Let's get going.
7:23
Let's get out of this archaic finance software and move toward
7:27
the future core memory and Brex.
7:31
SPEAKER_01: Okay.
7:31
SPEAKER_01: Yes.
7:31
SPEAKER_01: Yeah.
7:31
SPEAKER_01: I would, I would assume so.
7:32
SPEAKER_01: So I've, I've been to Jeff Lickman's lab at Harvard.
7:35
SPEAKER_01: where they have these brain slicers,
7:37
SPEAKER_01: which I think they were using on mice and human,
7:40
SPEAKER_01: small human samples at the time.
7:41
SPEAKER_01: But essentially with the connectome, you want to image this whole brain.
7:45
SPEAKER_01: We did the worm.
7:47
SPEAKER_01: We just did the fly.
7:48
SPEAKER_01: There's been many millions,
7:49
SPEAKER_01: if not,
7:50
I think billions at this point invested in trying to get a
7:52
human connectome.
7:54
SPEAKER_01: As far as I'm aware,
7:56
SPEAKER_01: I mean,
7:56
SPEAKER_01: we're on,
7:57
SPEAKER_01: to try to image a human brain,
7:59
SPEAKER_01: we're on like the cubic centimeter of a brain type levels.
8:04
SPEAKER_00: Well, certainly we haven't gotten billions.
8:06
SPEAKER_00: So how about I back up?
8:08
SPEAKER_00: Yes.
8:09
So,
8:09
you know,
8:10
there have been like a few milestones in this field of
8:13
connectomics.
8:14
SPEAKER_00: You know,
8:14
SPEAKER_00: at the very beginning,
8:15
SPEAKER_00: actually,
8:15
SPEAKER_00: at the very beginning of the 20th century,
8:17
SPEAKER_00: there was this guy,
8:18
SPEAKER_00: Santiago Román y Cajal.
8:20
SPEAKER_00: And he,
8:20
SPEAKER_00: along with Golgi,
8:21
SPEAKER_00: who you might have heard of,
8:22
SPEAKER_00: who developed this Golgi stain,
8:23
SPEAKER_00: figured out how to stain these neurons with silver.
8:27
SPEAKER_00: And it was like one neuron in like thousands.
8:31
And back then Cajal was looking at this with a light microscope
8:36
and drawing
8:37
SPEAKER_00: pictures of these neurons.
8:39
And you can look up these beautiful drawings from Cajal that
8:42
show you the shapes
8:42
SPEAKER_00: and sizes of all of these different neurons.
8:45
SPEAKER_00: And that was how neuroanatomy got kicked off.
8:48
SPEAKER_00: Sorry, roughly what year was this?
8:50
SPEAKER_00: Oh, this is like 1910-ish.
8:52
SPEAKER_00: Okay.
8:52
SPEAKER_00: Okay.
8:53
SPEAKER_00: Wow.
8:53
SPEAKER_00: Okay.
8:55
SPEAKER_00: And there were a couple of things that came out of that.
8:58
SPEAKER_00: There was this neuron doctrine,
9:00
SPEAKER_00: which was this articulation that the brain is made up of neurons,
9:03
SPEAKER_00: right?
9:03
SPEAKER_00: They're made up of these individual discrete units that compute.
9:08
SPEAKER_00: They talk at synapses, which are the connections between neurons.
9:11
SPEAKER_00: And that your brain has like 100 billion of these neurons.
9:17
SPEAKER_00: So that's like 1E11 in scientific notation.
9:19
SPEAKER_00: And we'll get to this, but that's what we call the company E11Bio.
9:22
One day we want to be able to build the tech that will let us
9:25
map all 1E11 neurons
9:27
SPEAKER_00: of a human connectome.
9:29
SPEAKER_00: But we are in many ways still very early.
9:32
SPEAKER_00: So like
9:33
SPEAKER_00: You mentioned there have been a few milestones over time.
9:37
First,
9:38
you could look at the shapes of the neurons,
9:40
but that didn't give you the circuit map.
9:43
These neurons are connected into circuits,
9:45
but you don't have a wiring diagram.
9:47
SPEAKER_00: And so then Sidney Brenner and his colleagues in the 70s are like,
9:50
SPEAKER_00: okay,
9:50
SPEAKER_00: let's try to get the wiring diagram of an entire animal,
9:54
SPEAKER_00: the C.
9:55
SPEAKER_00: elegans organism,
9:56
SPEAKER_00: this nematode.
9:57
SPEAKER_00: And so that was the proof of concept that it could be done.
10:01
SPEAKER_00: And we learned some things from that as a field.
10:04
And then,
10:04
like I mentioned,
10:05
there was this long winter where we're developing the tools.
10:09
There are pioneers who are trying to figure out how to get the
10:12
technology to do
10:13
SPEAKER_00: this in a scalable,
10:14
SPEAKER_00: automated way.
10:17
Winford Denk and Max Planck was developing this diamond-knife
10:21
slicing approach.
10:22
SPEAKER_00: Similarly,
10:23
SPEAKER_00: you mentioned Jeff Lichtman,
10:24
SPEAKER_00: his team was also developing this diamond-knife slicing approach.
10:28
And the fly was developed and mapped in very much the same way,
10:33
just at a larger scale.
10:35
SPEAKER_00: It was still sliced into these thousands of slices.
10:38
SPEAKER_00: Pictures were taken with electron microscopes of each one,
10:40
SPEAKER_00: and then they were stitched together into this 3D volume.
10:44
SPEAKER_00: And so the question is,
10:47
SPEAKER_00: If we wanna go further than the fly,
10:50
SPEAKER_00: if we wanna go to the mouse or eventually the human,
10:53
SPEAKER_00: because we haven't actually gotten to a mouse connectome yet,
10:56
SPEAKER_00: what do we need to do?
10:58
And is it gonna look like scaling up more electron microscopes
11:03
and better computer vision,
11:05
SPEAKER_00: Or is it going to look a little bit more like genomics?
11:07
And so this is where this training in George Church's lab and
11:11
other genomics
11:13
background is really interesting because things worked very
11:16
differently in
11:16
SPEAKER_00: genomics.
11:17
SPEAKER_00: The first big human genome project was completed with Sanger sequencing,
11:23
SPEAKER_00: this technology that was
11:25
SPEAKER_00: You know, very industrialized.
11:26
There were thousands of these Sanger sequencers running across
11:29
the country day in,
11:30
SPEAKER_00: day out to sequence the first human genome.
11:32
SPEAKER_00: And that cost, you know, billions of dollars.
11:34
SPEAKER_00: And,
11:34
SPEAKER_00: you know,
11:35
SPEAKER_00: Connectomics has not had something equivalent to,
11:37
SPEAKER_00: you know,
11:37
SPEAKER_00: a billion dollar human genome project.
11:39
SPEAKER_00: We're not there yet.
11:42
But around 2007,
11:45
things start getting really interesting because new technology
11:49
hits the scene.
11:50
And these microscopes,
11:53
a sequencer is actually a microscope in a box these days.
11:56
SPEAKER_00: It wasn't back then.
11:57
But around 2007,
11:58
Illumina and its competitors developed these next generation
12:01
sequencers.
12:02
And the cost of sequencing falls like four orders of magnitude
12:05
in like five years.
12:07
SPEAKER_00: It's this massive cost reduction.
12:09
And suddenly people are able to,
12:10
you know,
12:11
do sequencing in their own labs.
12:13
SPEAKER_00: They're able to buy these instruments.
12:15
SPEAKER_00: It's just massively cheaper.
12:17
SPEAKER_00: And so the question that,
12:18
SPEAKER_00: you know,
12:18
SPEAKER_00: E11 and I have been sort of obsessing over is,
12:22
SPEAKER_00: you know,
12:22
is there new technology that will let us have that next
12:25
generation sequencing
12:26
SPEAKER_00: moment for connectomics where,
12:28
SPEAKER_00: you know,
12:29
SPEAKER_00: we can start to conceive of what it would,
12:31
SPEAKER_00: you know,
12:31
SPEAKER_00: what it would take to go from the fly where we are today.
12:34
SPEAKER_00: to the mouse,
12:35
SPEAKER_00: which is 1,000 times bigger,
12:36
SPEAKER_00: or the human brain,
12:37
SPEAKER_00: which is 1,000 times bigger than that.
12:40
SPEAKER_01: That was fantastic context.
12:42
SPEAKER_01: And let's explain why it is so hard.
12:45
SPEAKER_01: You mentioned early on there's a
12:49
SPEAKER_01: There's a technique part.
12:51
SPEAKER_01: There's the machines.
12:52
SPEAKER_01: Some of this is physical labor and how you decide to mass produce that.
12:57
SPEAKER_01: You have to do these slices.
12:59
SPEAKER_01: And then part of this is computing power.
13:03
I mean,
13:04
I'm sure you'll tell us,
13:05
but we're talking about ridiculous amounts of data.
13:10
SPEAKER_01: So there's these different bottlenecks in this.
13:13
SPEAKER_01: What is the core...
13:17
SPEAKER_01: If you hear that somebody has done a fly and they're slicing the brain,
13:20
SPEAKER_01: when I first started to learn about this,
13:22
SPEAKER_01: I was like,
13:22
SPEAKER_01: oh,
13:23
SPEAKER_01: well,
13:23
why don't we just buy just tons of slicers then and just mass
13:28
produce this thing
13:29
SPEAKER_01: and just do the labor?
13:30
SPEAKER_01: Why, you know, what are the reasons that this has been so difficult?
13:35
SPEAKER_00: Yeah, I would say that there's two reasons why this is hard.
13:39
SPEAKER_00: The first one is,
13:41
and people have been running pilot projects to understand where
13:47
the bottlenecks
13:48
SPEAKER_00: actually are.
13:48
SPEAKER_00: And so there have been these amazing efforts
13:53
SPEAKER_00: So,
13:53
SPEAKER_00: for instance,
13:54
SPEAKER_00: you might have heard the Microns Project,
13:56
SPEAKER_00: which was an effort to map a cubic millimeter of mouse brain tissue.
13:59
SPEAKER_00: That's 1 500th of a mouse brain,
14:01
SPEAKER_00: but is the biggest volume of mouse brain that had been mapped to date.
14:07
SPEAKER_00: And that,
14:09
SPEAKER_00: along with other efforts,
14:11
SPEAKER_00: Harvard mapped a cubic millimeter of human tissue,
14:14
both of those efforts sort of illuminated where the current
14:17
bottleneck is,
14:19
SPEAKER_00: and it's tracing neurons through the images.
14:22
SPEAKER_00: So when you get these images,
14:24
SPEAKER_00: You have to slice by slice,
14:26
SPEAKER_00: cross-section by cross-section,
14:27
SPEAKER_00: trace the neurons from one sequential section to another.
14:32
SPEAKER_00: And this is like, again, it's the ultimate color by numbers problem.
14:35
SPEAKER_00: Instead of colored pencils, now you have AI.
14:37
But you still have to detect boundaries,
14:41
trace cross-sections,
14:42
and not make errors.
14:43
Because when you make an error,
14:45
then everything downstream of that,
14:47
the error propagates.
14:48
SPEAKER_00: The errors are correlated.
14:50
SPEAKER_01: Because it's like you're...
14:53
SPEAKER_01: I don't think this is, I'm gendering wires here.
14:57
SPEAKER_01: I think this is mostly a male problem,
14:59
is the bucket of wires that you have in your garage where
15:03
you've put all your old
15:05
cords and they're all tangled around each other and you're
15:08
trying to either
15:09
SPEAKER_01: detangle them or find the end of one.
15:12
SPEAKER_01: It's sort of like that, but just exponentially worse.
15:15
SPEAKER_00: Yeah, I mean, it's a rat's nest.
15:16
SPEAKER_00: Yeah, you can think of the rat's nest of wires as a good analogy here.
15:19
SPEAKER_00: Yeah.
15:21
And unlike the rat's nest where you can sort of follow it with
15:23
your fingertips
15:24
SPEAKER_00: here,
15:24
SPEAKER_00: all you have to work with is vision.
15:27
SPEAKER_00: And so it turns out that this is the bottleneck.
15:31
The Wellcome Trust released a report a couple of years ago
15:33
where they sort of gamed
15:34
out what would it take to scale up electron microscopes to the
15:37
mouse connectome.
15:39
SPEAKER_00: And the conclusion is, at least at that time,
15:43
tracing the neurons would be 98% of the cost of mapping the
15:48
mouse connectome.
15:49
SPEAKER_00: Even for the fly,
15:50
and not to detract in any way from the fly as a massive
15:54
accomplishment and
15:54
SPEAKER_00: milestone,
15:56
SPEAKER_00: first you use automatic computer vision to do a first draft of tracing,
16:01
SPEAKER_00: but there's then a bunch of errors.
16:02
And so then what the Princeton team did was they got together
16:07
human annotators to
16:09
SPEAKER_00: manually correct the outputs of the computer vision models.
16:12
And that was like 33 person years of manual annotation and
16:15
correction to get to
16:16
SPEAKER_00: that first fly connecting.
16:18
SPEAKER_00: So tracing is the bottleneck.
16:20
SPEAKER_00: And that's the problem that the field is currently grappling with.
16:23
SPEAKER_00: Were those poor grad students or mechanical?
16:26
SPEAKER_00: Oh,
16:27
SPEAKER_00: you know,
16:27
SPEAKER_00: I mean,
16:27
SPEAKER_00: it,
16:28
SPEAKER_00: it was a collection,
16:28
SPEAKER_00: you know,
16:28
SPEAKER_00: there are citizen scientists,
16:30
SPEAKER_00: there were,
16:30
SPEAKER_00: you know,
16:30
SPEAKER_00: professional annotators,
16:32
SPEAKER_00: you know,
16:32
SPEAKER_00: there,
16:32
SPEAKER_00: there exists a occupation,
16:34
SPEAKER_00: which is,
16:34
SPEAKER_00: you know,
16:34
SPEAKER_00: a professional,
16:35
SPEAKER_00: you know,
16:35
SPEAKER_00: neuroscientific annotator.
16:37
SPEAKER_00: So,
16:38
SPEAKER_00: you know,
16:39
SPEAKER_00: um,
16:40
SPEAKER_00: and it's,
16:40
SPEAKER_00: it's,
16:40
SPEAKER_00: it gets the job done,
16:42
SPEAKER_00: but it's,
16:43
SPEAKER_00: you know,
16:44
SPEAKER_00: expensive and it requires human input.
16:45
And so,
16:46
you know,
16:47
there are a couple of different ways to attack the problem.
16:49
SPEAKER_00: Many people are attacking it with AI.
16:51
SPEAKER_00: you know, as you get better AI, then you can bring these costs down.
16:54
SPEAKER_00: But,
16:55
SPEAKER_00: you know,
16:55
SPEAKER_00: the cost of tracing has to fall by at least a hundredfold for it to be,
16:59
SPEAKER_00: you know,
16:59
SPEAKER_00: a equal cost driver,
17:02
SPEAKER_00: you know,
17:02
SPEAKER_00: with imaging and,
17:03
SPEAKER_00: you know,
17:04
SPEAKER_00: other pre-processing and then tracing.
17:06
SPEAKER_00: And it needs to fall by a thousandfold if you want to,
17:07
SPEAKER_00: you know,
17:09
SPEAKER_00: have it be a negligible cost driver.
17:10
SPEAKER_00: So this is the bottleneck that needs to be solved.
17:13
SPEAKER_01: Okay,
17:13
SPEAKER_01: and then before we get to different approaches to solving this problem,
17:18
SPEAKER_01: let's explain why you would even want this in the first place.
17:21
And I almost feel like there's a pre-AI answer and possibly a
17:27
post-AI answer
17:28
because people were chasing the connectome before this recent
17:31
takeoff in AI
17:33
SPEAKER_01: technology.
17:34
SPEAKER_01: And most of the interest was that, look...
17:38
SPEAKER_01: The brain defines everything about us.
17:41
SPEAKER_01: We should understand how it works.
17:43
SPEAKER_01: We know precious little about its mechanics.
17:46
SPEAKER_01: So let's build this wiring diagram.
17:49
Let's just map a whole thing so we can at least understand one
17:54
brain and how
17:55
SPEAKER_01: everything is interconnected.
17:57
And hopefully we will come to new insights about how this organ
18:02
works by doing this.
18:03
SPEAKER_01: It's a slightly controversial thing.
18:06
SPEAKER_01: thought,
18:07
SPEAKER_01: I think,
18:07
because there's some people that would argue just having that
18:10
map doesn't do a lot
18:11
SPEAKER_01: for you.
18:12
And you're almost definitely doing this on a brain in a moment
18:16
of time that is dead.
18:18
SPEAKER_01: You don't get to see what's happening actively in this brain.
18:22
And so there's been some debate about that,
18:24
whether this is like a worthwhile exercise.
18:26
SPEAKER_01: I tend to fall.
18:27
SPEAKER_01: It seems obvious to me that we should
18:30
map this thing and understand how it works and then i feel like
18:33
and i know you're
18:34
into this and i want to get into it you know i think there's
18:36
this post ai answer
18:38
where it's it's like well okay now we're making these very
18:43
powerful artificial
18:44
minds that are loosely based on this neuronal structure that we
18:50
have in our heads
18:51
and if if we want these two things to one day maybe be aligned
18:55
and and and
18:57
SPEAKER_01: not maybe mimicking each other,
18:59
SPEAKER_01: but borrowing from each other enough that we're not,
19:01
SPEAKER_01: not like destroying,
19:03
SPEAKER_01: um,
19:04
SPEAKER_01: the AI isn't destroying us.
19:05
SPEAKER_01: There's some kind of, um, humanness in it.
19:08
SPEAKER_01: You know,
19:08
SPEAKER_01: you would,
19:09
you would want to replicate that by having this model of the
19:12
brain somehow injected
19:14
SPEAKER_01: into these AI systems.
19:15
SPEAKER_01: But what's your, like, why did you get into this?
19:17
SPEAKER_01: Why do you think it's important?
19:19
SPEAKER_00: Yeah.
19:20
SPEAKER_00: I mean,
19:21
SPEAKER_00: you know,
19:21
SPEAKER_00: if you're asking me why I got into that,
19:23
SPEAKER_00: you know,
19:23
SPEAKER_00: maybe we can just like go back to the beginning for just a second.
19:25
I thought because this could go back to your childhood or I
19:28
don't know where,
19:29
SPEAKER_01: where this would start.
19:30
SPEAKER_00: Yeah.
19:30
SPEAKER_00: I mean, you know, I'm an optimist.
19:32
SPEAKER_00: Um, like I, I grew up like, uh,
19:35
on golden age science fiction,
19:37
I was really drawn to positive visions of the future.
19:42
SPEAKER_00: One day when I was a kid,
19:43
the school bus was late and I was very upset because I was
19:45
going to miss Star Trek.
19:46
SPEAKER_00: So that was what shaped my worldview.
19:50
SPEAKER_00: And then later on,
19:52
SPEAKER_00: I had conversations with my uncle,
19:56
SPEAKER_00: I realized that maybe I could do something.
19:57
SPEAKER_00: He was an engineer and he convinced me to be an engineer.
20:01
SPEAKER_00: And
20:04
SPEAKER_00: Like I mentioned, I went to the University of Toronto to study there.
20:08
SPEAKER_00: And the reason I got into this was that I read this essay.
20:14
I was studying physics engineering science,
20:16
which was a rebrand of engineering physics.
20:18
SPEAKER_00: And I read this essay by Feynman,
20:19
SPEAKER_00: which was like,
20:19
SPEAKER_00: it's really easy to solve these biological problems.
20:22
SPEAKER_00: if you can just look at the thing, then you'll be able to answer.
20:25
SPEAKER_00: You'll be able to see the sequence of bases in the chain.
20:28
SPEAKER_00: And I'm like, oh, actually, this is a great approach to biology.
20:31
SPEAKER_00: This is what we should be doing.
20:32
SPEAKER_00: We should be building microscopes.
20:34
SPEAKER_00: And this was really important to me because when I was 12,
20:37
SPEAKER_00: I started experiencing chronic pain and arthritis for the first time.
20:40
SPEAKER_00: And I spent my fair share of time at SickKids Hospital.
20:43
SPEAKER_00: I spent a lot of time with my physical therapist.
20:47
SPEAKER_00: And
20:49
SPEAKER_00: In the process,
20:50
I learned a lot about chronic pain and learned that actually
20:54
there's lots of people
20:55
SPEAKER_00: in the world who have brain disorders.
20:57
Actually,
20:57
one in three people in the world at some point will experience
21:00
a brain disorder.
21:02
And there are different ways to try to attack that problem and
21:07
try to reduce that
21:08
SPEAKER_00: level of suffering.
21:10
SPEAKER_00: For me, in the late 2000s, it was like genomics.
21:14
SPEAKER_00: But over time,
21:15
I think a lot of people have realized that there's this
21:17
potential in neurotech to
21:20
SPEAKER_00: really take on some of these serious mental health,
21:27
SPEAKER_00: brain disorders,
21:28
SPEAKER_00: and also some of the other stuff that you just mentioned.
21:30
SPEAKER_00: We'll get to that.
21:32
SPEAKER_00: And that is like.
21:37
SPEAKER_00: that realization is what got me into connectomics.
21:41
I conceptualized chronic pain as like a connectopathy,
21:46
like a disorder of the wiring.
21:48
SPEAKER_00: And that's,
21:49
SPEAKER_00: I think,
21:49
a good way to think about some of the utility of being able to
21:52
map these brain
21:54
SPEAKER_00: circuits at the level of synaptic connections.
21:56
SPEAKER_00: And we haven't actually defined connectome for the audience,
21:58
SPEAKER_00: but the connectome is just the set,
22:00
SPEAKER_00: the wiring diagram of the entire brain.
22:03
SPEAKER_00: And you can,
22:03
SPEAKER_00: you know,
22:03
SPEAKER_00: you can map an entire brain,
22:04
SPEAKER_00: you can map,
22:05
SPEAKER_00: you know,
22:05
SPEAKER_00: smaller connectomes within brains.
22:07
SPEAKER_00: But,
22:08
SPEAKER_00: you know,
22:08
the idea is that if you can actually look at the changes in the
22:11
wiring between like
22:12
SPEAKER_00: a healthy brain and a disordered brain,
22:15
SPEAKER_00: that will give you a new target that you'd be able to intervene with,
22:21
with a drug or a brain computer interface in order to then
22:25
restore function and
22:26
SPEAKER_00: health to the patient.
22:28
SPEAKER_00: So in that sense,
22:30
SPEAKER_01: because sometimes I find the comparison with the genome unsatisfying,
22:34
SPEAKER_01: but in that sense,
22:34
SPEAKER_01: somewhat similar in that you could take a healthy person's genome,
22:40
SPEAKER_01: somebody with a specific type of cancer,
22:42
SPEAKER_01: and if you overlay those two sequences,
22:44
SPEAKER_01: not that it usually works out this way,
22:46
SPEAKER_01: but you happen to see this one bit of DNA that's quite different,
22:52
SPEAKER_01: and then that's maybe a place to go looking for an underlying cause.
22:57
SPEAKER_00: Yeah, I mean, the analogy works in a couple of places.
23:00
SPEAKER_00: So,
23:00
SPEAKER_00: you know,
23:01
SPEAKER_00: the first thing you need in genomics is one genome,
23:04
SPEAKER_00: and that helps you get oriented,
23:05
SPEAKER_00: right?
23:06
SPEAKER_00: And so,
23:06
SPEAKER_00: you know,
23:07
SPEAKER_00: as we get these connectomes,
23:09
SPEAKER_00: these reference connectomes,
23:10
SPEAKER_00: that will really help us understand,
23:11
SPEAKER_00: you know,
23:11
SPEAKER_00: what
23:12
SPEAKER_00: normal or healthy wiring looks like.
23:15
But the real value of genomics was when sequencing got cheap
23:18
enough that you were
23:19
SPEAKER_00: able to map many genomes across many different conditions,
23:25
SPEAKER_00: disease models and developmental models and so on.
23:27
SPEAKER_00: And then you could start looking at the variation between individuals.
23:31
SPEAKER_00: And as you do that,
23:33
you then can track down what are the differences and that
23:36
creates hypotheses that
23:38
SPEAKER_00: you can then pursue with drugs.
23:41
SPEAKER_00: And it's going to be similar for connectomics.
23:44
SPEAKER_00: As we develop better tools that reduce the cost,
23:48
SPEAKER_00: and this is why E11 and I am relentlessly focused on cost,
23:52
SPEAKER_00: if we can make it so cheap that we can do many connectomes,
23:55
then we'll be able to look at these different conditions and
23:58
start to diagnose what
23:59
SPEAKER_00: are the connectopathies,
24:00
which psychiatric disorders are correctly understood as
24:04
connectopathies and which
24:06
SPEAKER_00: might not be.
24:07
SPEAKER_01: I guess it's very difficult because we don't know what worms suffer from,
24:11
SPEAKER_01: but were we able to do anything along those lines?
24:15
SPEAKER_01: How many,
24:16
SPEAKER_01: I don't know how many worm connectomes,
24:17
did we just do the one and then we sort of stopped or do we
24:20
have multiple to
24:21
SPEAKER_01: compare against?
24:23
SPEAKER_00: We do now,
24:23
SPEAKER_00: but there was definitely,
24:25
SPEAKER_00: you know,
24:26
SPEAKER_00: after,
24:26
even the first one was actually sort of a composite of several
24:29
different worms
24:30
because it was difficult to get the slicing technology working
24:32
well enough to be
24:33
SPEAKER_00: able to do an entire worm in one go.
24:35
But this is why it's important to be able to get connectomes
24:39
from disease models
24:42
SPEAKER_00: themselves or developmental models,
24:43
because the worm is just not a good model for the mental health
24:48
disorders or other
24:49
SPEAKER_00: disorders that people live with.
24:51
SPEAKER_00: So the worm and the fly, maybe not.
24:54
SPEAKER_00: But if you can start talking about the mouse,
24:56
SPEAKER_00: where we actually do have some good models for diseases,
25:00
SPEAKER_00: then you get really interesting.
25:02
SPEAKER_00: And there are other things where connectomics might be useful as well,
25:08
SPEAKER_00: not just mice,
25:09
SPEAKER_00: but there are organoid models,
25:10
SPEAKER_00: which are good models for disease.
25:13
There are some cases where you might be able to look at
25:15
postmortem tissue and be
25:16
SPEAKER_00: able to look at that.
25:17
So I think there are many different cases where once we can get
25:21
connectomics
25:22
SPEAKER_00: working in mammals,
25:23
then it's going to be very transformative for therapeutics and
25:27
for the brain.
25:29
SPEAKER_00: And like for the naysayers,
25:33
the people who've been skeptical that we learn that much just
25:35
because we get a map,
25:37
SPEAKER_01: it's kind of funny because I think when the genome was first sequenced,
25:40
SPEAKER_01: a lot of people thought,
25:42
SPEAKER_01: oh,
25:42
SPEAKER_01: we've done this once and the next day.
25:44
SPEAKER_01: And you wake up and, I mean, it really was, I lived through this.
25:48
SPEAKER_01: I mean,
25:48
SPEAKER_01: I'm simplifying things a little bit,
25:50
SPEAKER_01: but it was sort of like,
25:50
SPEAKER_01: oh,
25:51
you do it and then we're going to just cure all kinds of stuff
25:54
kind of in the next
25:56
SPEAKER_01: two,
25:56
SPEAKER_01: three,
25:56
SPEAKER_01: four years just by looking at this one genome.
25:59
SPEAKER_01: And then it turned out to be exactly like you said.
26:02
SPEAKER_01: We need a lot of this.
26:03
SPEAKER_01: It's been, what, 25?
26:05
SPEAKER_01: years now of,
26:06
SPEAKER_01: of,
26:07
SPEAKER_01: and we have made progress,
26:08
SPEAKER_01: but,
26:09
SPEAKER_01: but still I think tip of the iceberg of where all this could go.
26:13
SPEAKER_01: But,
26:13
SPEAKER_01: but for the people who,
26:16
SPEAKER_01: I remember when I was doing a lot of connectomic stories,
26:18
SPEAKER_01: there was definitely very bright people who were just like,
26:21
SPEAKER_01: yeah,
26:21
SPEAKER_01: this is,
26:21
SPEAKER_01: this is like a,
26:22
SPEAKER_01: this is not the best way we should spend our money.
26:24
But are they,
26:25
do you think they were solely in the,
26:29
we're just getting one brain camp?
26:31
SPEAKER_01: Is that the central argument?
26:32
SPEAKER_01: Does, does anybody debate that if we had,
26:35
lots of brain images that we would learn something from it that
26:38
seems like a hard
26:39
thing to to go against to me i mean there are always debates
26:42
over resource
26:43
allocation but you know i i think um attacking you know we're
26:48
gonna get one brain
26:49
and stop is a bit of a straw man you know it's it's really
26:52
about bending the cost
26:53
curve so it becomes so cheap to get this kind of data that you
26:57
know then you don't
26:57
SPEAKER_00: need to have the debate
26:59
And so I think that's what's really interesting around some of
27:02
the new approaches
27:03
where we're focusing on reducing the cost because then we can
27:06
short circuit the
27:07
SPEAKER_00: debate.
27:08
SPEAKER_00: But I think you're right.
27:08
SPEAKER_00: I mean, you have to set expectations in the right place.
27:12
It took genomics many years to develop drugs because it takes a
27:16
long time to make
27:18
SPEAKER_00: these drugs and to
27:20
SPEAKER_00: and to understand what are good targets.
27:22
SPEAKER_00: I think a good question is, what are the alternatives?
27:25
Some of these disorders are tractable using genomics tools,
27:30
but many others are very polygenic.
27:33
SPEAKER_00: It's like when there are many genes,
27:35
SPEAKER_00: there's something like 30 genes,
27:36
SPEAKER_00: for instance,
27:37
SPEAKER_00: implicated in schizophrenia.
27:38
SPEAKER_00: Many of them are synaptic,
27:39
but there's no one target that is like an obviously good target
27:45
for intervention.
27:45
SPEAKER_00: And so I think it is worth asking,
27:47
can we get something new and can we test the hypothesis that
27:51
maybe if you look at
27:53
SPEAKER_00: the circuits as the origin of the connectopathy,
27:57
SPEAKER_00: if it is a connectopathy,
27:59
SPEAKER_00: then you might be able to make more progress in this area.
28:02
SPEAKER_01: I see.
28:02
SPEAKER_01: So some people would argue that
28:05
SPEAKER_01: We solve these issues just through DNA,
28:08
SPEAKER_01: and that's going to be the major way to get at this stuff.
28:11
SPEAKER_01: And the circuitry is revealing about some underlying cause for something.
28:17
SPEAKER_00: Yeah,
28:17
SPEAKER_00: I think there are some,
28:18
SPEAKER_00: like Huntington's,
28:19
SPEAKER_00: where it's like a very clearly monogenic,
28:22
SPEAKER_00: there's a single gene,
28:22
SPEAKER_00: you know what the gene is.
28:25
SPEAKER_00: That's less likely to be impacted by connectomics.
28:28
But for places where we don't really have good hypotheses or
28:33
targets that have
28:34
SPEAKER_00: panned out anyway,
28:36
SPEAKER_00: having new tools in the toolbox I think is going to be really important.
28:39
SPEAKER_01: Well,
28:39
and there's this—I want to get into what E11 does and how you
28:43
guys are going about
28:43
SPEAKER_01: this,
28:43
SPEAKER_01: so I run the risk of sending us on a slight tangent really quick.
28:46
SPEAKER_01: But there's a frustrating part of this to me,
28:49
SPEAKER_01: which is whether—no matter which way you want to debate it,
28:52
it just seems so obvious to me that we should know as much as
28:55
possible about the
28:56
SPEAKER_01: thing that defines our reality.
29:00
SPEAKER_01: And I—it's—
29:02
SPEAKER_01: I know people have spent centuries studying their brains.
29:04
SPEAKER_01: It's frustrating to me that sometimes it feels like second fiddle.
29:10
I understand why we are spending what must be now trillions of
29:15
dollars creating
29:16
SPEAKER_01: artificial brains.
29:18
However,
29:21
because of all the economic incentives behind that,
29:24
I find it baffling that...
29:27
The world hasn't pumped more money into brain research
29:31
historically when it seems so important.
29:34
SPEAKER_00: Well, you know, I think there's a lot of exciting science to do.
29:39
SPEAKER_00: And so,
29:39
SPEAKER_00: you know,
29:39
SPEAKER_00: I can understand why many different projects are,
29:43
SPEAKER_00: you know,
29:44
SPEAKER_00: competitive and interesting.
29:46
SPEAKER_00: But I think something that has changed recently that is like, you
29:52
observing and then re-evaluating connectomics in its light is
29:54
the fact that
29:55
SPEAKER_00: neurotech is changing.
29:57
SPEAKER_00: In the last few years,
29:59
SPEAKER_00: our ability to interface with the brain using BCIs,
30:02
SPEAKER_00: and you've covered a bunch of them recently,
30:04
SPEAKER_00: there's this proliferation of techniques to get into the brain.
30:07
SPEAKER_00: And I think we're getting to the point now
30:09
where our ability to interface with the brain might actually be
30:12
outstripping our
30:13
SPEAKER_00: ability to know what we're interfacing with.
30:16
SPEAKER_00: And so now there's a good argument,
30:18
SPEAKER_00: I think,
30:19
SPEAKER_00: or another good argument for why we want to map out these circuits,
30:22
SPEAKER_00: because now we have the tools where if we can...
30:26
SPEAKER_00: I talked about drugs,
30:26
but brain-computer interfaces are good treatments for brain
30:30
disorders as well.
30:31
If we know where we should put the electrodes or which circuits
30:35
to stimulate,
30:36
SPEAKER_00: that could be very transformative.
30:39
SPEAKER_01: Okay,
30:40
SPEAKER_01: since day one,
30:40
the Core Memory podcast has been supported by the fine people
30:44
at E1 Ventures.
30:45
SPEAKER_01: They are a young and ambitious VC firm in Silicon Valley,
30:49
SPEAKER_01: investing in young and ambitious companies and people.
30:53
SPEAKER_01: Thank you so much to E1 Ventures for all your support.
30:57
SPEAKER_01: Right.
30:57
SPEAKER_01: Okay.
30:57
SPEAKER_01: And then even like some basic stuff,
30:59
I remember sitting with Jeff Lickman and I think he was
31:02
pointing me to some part of
31:03
SPEAKER_01: an axon and there was this turban-like shape and he was like,
31:09
SPEAKER_01: we just,
31:09
SPEAKER_01: we didn't even know these things existed before we did this.
31:13
SPEAKER_01: So,
31:13
SPEAKER_01: I mean,
31:14
SPEAKER_01: that just jumped out at me as,
31:15
SPEAKER_01: well,
31:15
SPEAKER_01: if there's parts of the bray,
31:16
SPEAKER_01: we didn't even,
31:18
we didn't even know this thing lurked and he found them all
31:22
over the place on this
31:23
SPEAKER_01: tiny sample.
31:24
SPEAKER_01: Yeah.
31:24
there's no doubt we're going to learn a lot of new biology just
31:28
along the way.
31:29
SPEAKER_00: If you look at this Microns project,
31:31
some of the most active users of that mouse cubic millimeter
31:34
are cell biologists
31:35
SPEAKER_00: who are interested in glial cells.
31:36
They're not actually there for the connectomics,
31:39
they're there for the cell biology.
31:41
SPEAKER_00: So
31:42
SPEAKER_00: But this happened in genomics as well.
31:44
SPEAKER_00: You build DNA sequencers because you want to sequence DNA.
31:47
And then fast forward 10 years and people are using them to
31:50
sequence RNA for cell typing.
31:52
SPEAKER_00: They're using them as diagnostic tools.
31:56
SPEAKER_00: They're even using them for data storage, for IT, right?
31:59
SPEAKER_00: How do you want to archive your data for the long run?
32:01
SPEAKER_00: You turn it into DNA and then bury it, right?
32:03
SPEAKER_00: Yeah.
32:04
I think these general purpose tools like close to the root of
32:08
the tech tree like
32:09
SPEAKER_00: genomics or kinetomics,
32:10
I think have many unexpected applications as you make them
32:16
accessible and cheap.
32:17
SPEAKER_01: Okay,
32:17
SPEAKER_01: well,
32:18
SPEAKER_01: I will not embarrass myself fully trying to explain what E11 does,
32:23
SPEAKER_01: and thankfully I don't have to,
32:24
SPEAKER_01: but loosely,
32:25
SPEAKER_01: just to get us going,
32:27
SPEAKER_01: all the stuff we've been talking about,
32:29
you guys have a new technique to try to image the brain that
32:33
gets around some of
32:34
SPEAKER_01: these limitations.
32:36
SPEAKER_01: that we've talked about.
32:37
SPEAKER_01: And like you mentioned, you're very focused on cost.
32:40
SPEAKER_01: And so this is, um, just a major, major issue.
32:44
SPEAKER_01: And so,
32:45
SPEAKER_01: yeah,
32:45
SPEAKER_01: I mean,
32:45
SPEAKER_01: you guys are taking a swing that,
32:47
SPEAKER_01: that other people,
32:47
SPEAKER_01: um,
32:49
SPEAKER_01: have not give us the like 30,000 foot view of,
32:55
SPEAKER_01: of how this technology works.
32:56
SPEAKER_00: Yeah.
32:57
SPEAKER_00: So, you know, E11Bio is a nonprofit research org.
33:01
SPEAKER_00: It's a focused research organization, and we can talk about that.
33:05
SPEAKER_00: And the purpose of E11 is to,
33:07
SPEAKER_00: you know,
33:08
SPEAKER_00: develop these new tools that will make it possible,
33:11
SPEAKER_00: tractable,
33:12
SPEAKER_00: and
33:13
SPEAKER_00: inexpensive to map mammalian connectomes.
33:15
So we're really focused on this problem of how do we go to the
33:18
next level,
33:19
how to create this next generation sequencing moment for the
33:23
mammalian brain as we
33:24
SPEAKER_00: go from the fly where we are today to where we want to go,
33:26
SPEAKER_00: which is the mouse,
33:27
SPEAKER_00: and then ultimately human brains.
33:30
SPEAKER_00: And the idea that our thesis is not a new idea.
33:38
SPEAKER_00: It's been around since the mid-2000s.
33:42
SPEAKER_00: The idea is to make every neuron in the brain a different color.
33:47
SPEAKER_00: And if you go back to your rat's nest of wires in the garage,
33:50
SPEAKER_00: if they're all the same color,
33:51
SPEAKER_00: then it's going to be really hard to untangle them.
33:53
SPEAKER_00: But if they're different colors,
33:55
SPEAKER_00: then it's suddenly much easier to untangle that big ball of wire.
33:59
SPEAKER_01: So when we usually image these brains, I mean, it's just tissue.
34:03
SPEAKER_01: It's all coming back.
34:05
SPEAKER_00: The wires look the same.
34:07
SPEAKER_00: Yeah.
34:07
I mean,
34:07
the state of the art is you use electron microscopy,
34:09
which is typically black and white.
34:12
SPEAKER_00: And,
34:13
SPEAKER_00: you know,
34:13
you get these grayscale images out and then you have to trace
34:15
neurons in grayscale.
34:18
And when you use light microscopes and you use genetic
34:23
engineering,
34:24
you can make every neuron
34:27
a different color by delivering combinations of fluorescent
34:31
proteins to the neurons
34:32
SPEAKER_00: in the brain.
34:33
SPEAKER_00: And, you know, this idea was first implemented as Brainbow.
34:37
This was actually,
34:38
you know,
34:38
another technology out of Jeff Lichtman's lab.
34:40
SPEAKER_00: Great name.
34:40
SPEAKER_00: And, you know, Google it.
34:42
SPEAKER_00: It's beautiful.
34:43
SPEAKER_00: I mean, it creates, you know, these beautiful images.
34:45
SPEAKER_00: The images are, people should Google it.
34:48
SPEAKER_00: It's awesome.
34:49
But,
34:49
you know,
34:49
the challenge with,
34:50
and so this was one of the angles of attack on connectomics,
34:54
SPEAKER_00: But the challenge,
34:56
and this is the same challenge that Cajal had back at the dawn
35:00
of neuroanatomy,
35:01
SPEAKER_00: is that you can't generally see the connections between neurons,
35:05
SPEAKER_00: or at least not in a dense,
35:07
SPEAKER_00: efficient way,
35:08
SPEAKER_00: in light microscopes.
35:10
SPEAKER_00: And therefore, you can't get the connectome out.
35:13
SPEAKER_00: And so you can make these neurons different colors, but the
35:16
SPEAKER_00: the wires themselves overlap because a wavelength of light is big,
35:21
it's like 500 nanometers or more depending on where you are in
35:23
the spectrum,
35:24
SPEAKER_00: and the cleft between neurons is tens of nanometers at a synapse.
35:28
So it all blurred together and the state of the art ended up
35:32
becoming electron
35:33
microscopy because that was the only method with the resolution
35:36
necessary to see
35:38
SPEAKER_00: these connections between cells.
35:40
SPEAKER_00: But then something changes.
35:41
SPEAKER_00: I did my PhD in Ed Boyden's lab,
35:45
SPEAKER_00: and around 2015,
35:45
SPEAKER_00: Fei Chen and Paul Tilburg and Ed Boyden develop expansion microscopy.
35:51
SPEAKER_00: And that is one of the game changers.
35:54
SPEAKER_00: So I'm going to call it EXM expansion microscopy.
35:56
SPEAKER_00: And you essentially embed the brain in a swellable polymer,
36:00
SPEAKER_00: and then you add water,
36:02
SPEAKER_00: it swells,
36:03
SPEAKER_00: it makes the brain physically bigger.
36:06
SPEAKER_00: These days, you can make it 20 times bigger in every dimension.
36:09
SPEAKER_00: And that physically moves apart the neurons.
36:13
And so now two wires that were overlapping that you couldn't
36:17
resolve in a light
36:18
SPEAKER_00: microscope now can be resolved.
36:20
SPEAKER_01: So you're literally taking one of these samples and stretching it out.
36:24
SPEAKER_01: And I was just with Ed Boyden
36:27
SPEAKER_01: I didn't know this because immediately I was like,
36:31
SPEAKER_01: oh,
36:31
SPEAKER_01: well,
36:31
SPEAKER_01: you're going to get some crazy distortion when you start doing that.
36:34
SPEAKER_01: But I mean,
36:34
SPEAKER_01: he was telling me it's like still,
36:37
SPEAKER_01: I mean,
36:37
SPEAKER_01: they're finding it's like 99% holding scale and everything.
36:39
SPEAKER_01: It preserves the structure remarkably well.
36:41
SPEAKER_00: Even version 1.0 preserves the structure remarkably well.
36:48
SPEAKER_00: you're doing a little bit of a balancing act.
36:50
You're digesting some of the structural proteins enough so that
36:53
instead of
36:54
SPEAKER_00: stretching it,
36:55
SPEAKER_00: you're just moving it apart.
36:56
SPEAKER_00: You're moving it apart isotropically,
36:58
SPEAKER_00: which just means everything,
36:59
SPEAKER_00: all the biomolecules move the same distance apart from each other.
37:02
SPEAKER_00: And over the last 10 years, this has been very effectively optimized.
37:06
And so now it really sort of preserves this 3D structure,
37:10
which is very important.
37:10
SPEAKER_00: Which is crazy, right?
37:13
SPEAKER_01: So you were in the lab when this was,
37:15
SPEAKER_01: were you in the lab when this was taking place or being discovered?
37:18
SPEAKER_01: Just after.
37:19
SPEAKER_01: Okay.
37:19
SPEAKER_01: Okay.
37:20
SPEAKER_01: I mean, people must've been freaking out.
37:22
SPEAKER_01: I mean, it, it launched a field.
37:24
SPEAKER_01: Yeah.
37:25
SPEAKER_01: Yeah.
37:25
SPEAKER_01: Yeah.
37:25
SPEAKER_01: I mean, just it boggles my mind that this works, but.
37:31
SPEAKER_01: Yeah.
37:31
SPEAKER_01: I mean,
37:31
SPEAKER_01: Ed described it as like the same,
37:33
I don't know if he was saying it was the same material as what
37:36
goes in a baby
37:37
SPEAKER_01: diaper or?
37:37
SPEAKER_00: Yeah.
37:37
Or like,
37:38
you know,
37:38
there's dinosaurs that you would put in the bathtub that would
37:40
expand.
37:40
SPEAKER_00: Right.
37:41
Actually,
37:41
I'm not sure that that's the same material,
37:42
but it's the same concept.
37:43
It's like,
37:44
you know,
37:44
you add water and then,
37:45
you know,
37:46
the material expands.
37:47
SPEAKER_01: Okay,
37:47
SPEAKER_01: so you're studying at MIT,
37:49
SPEAKER_01: you're seeing the development of this technology take hold,
37:53
SPEAKER_01: and I assume this starts to get you thinking about...
37:56
SPEAKER_00: Yeah,
37:57
SPEAKER_00: so I met,
37:59
SPEAKER_00: you know,
38:00
SPEAKER_00: I started in the lab,
38:01
SPEAKER_00: I'm very interested in,
38:02
SPEAKER_00: you know,
38:02
SPEAKER_00: how do we apply this new tool expansion to Connectomics?
38:06
And I'm also inspired by this essay that Tony Zador wrote in
38:10
2012 called Sequencing
38:10
SPEAKER_00: the Connectome.
38:11
And it was an articulation of this idea that if you can make
38:15
neurons different
38:15
SPEAKER_00: colors or even better,
38:17
SPEAKER_00: give them different molecular ID tags,
38:20
you can distinguish them from each other and you can use many
38:22
of the tools that we
38:23
SPEAKER_00: have.
38:24
SPEAKER_00: in sequencing and then apply them in order to,
38:27
SPEAKER_00: you know,
38:28
SPEAKER_00: map the brain more effectively by using these,
38:32
SPEAKER_00: they're called cell barcodes.
38:33
SPEAKER_00: Barcode is just a term of art.
38:34
SPEAKER_00: It just means, you know, the cell now has a molecular fingerprint.
38:37
SPEAKER_00: Like a tag, yeah.
38:37
SPEAKER_00: And you use the tag to correct errors in the tracing.
38:40
And so I spent a couple of years working on that,
38:44
and there are some challenges.
38:46
It's difficult to get the barcodes to express throughout the
38:50
cell uniformly.
38:52
When you implement them in RNA,
38:53
sometimes they don't make it down the axon.
38:55
SPEAKER_00: Other times they don't make it up the dendrites.
38:57
SPEAKER_00: These are the two different branches of the neuron.
39:00
SPEAKER_00: And so this is a challenge.
39:03
SPEAKER_00: So the idea is nice, but the implementation is difficult.
39:06
SPEAKER_00: This brainbow technique is another thing that I look at.
39:09
SPEAKER_00: But the challenge with brainbow,
39:10
SPEAKER_00: even when combined with expansion microscopy,
39:12
is that there's only about a hundred or so distinguishable hues
39:17
in brainbow.
39:17
SPEAKER_00: You have red,
39:18
SPEAKER_00: green,
39:18
SPEAKER_00: and blue fluorescent proteins,
39:20
SPEAKER_00: and they can be expressed at different levels.
39:24
But 100 is much,
39:25
much less than the 100 million neurons in a mouse brain.
39:29
SPEAKER_00: So it's not enough.
39:31
SPEAKER_00: And after a couple of years, I'm like, man, this is hard.
39:36
SPEAKER_00: This is going to need a team of 10 or 15 people in five years.
39:39
SPEAKER_00: I'm not going to be able to do this as a graduate student on my own.
39:42
SPEAKER_00: And I spent some time working on comparatively simple stuff,
39:45
SPEAKER_00: sequencing genomes inside intact cells with microscopes,
39:50
SPEAKER_00: looking at the folds of the 3D structure of the genome.
39:53
SPEAKER_00: This is actually useful tech that we're now implementing at E11,
39:56
SPEAKER_00: but I had to take a break from the problem.
39:59
And around 2019,
40:02
Adam Marblestone,
40:03
Sam Rodriguez,
40:04
and I all overlapped in Ed's lab.
40:06
SPEAKER_00: And we asked this question, wait a second, what if we did
40:11
SPEAKER_00: have an organization where you had 10 or 15 people working on this?
40:15
SPEAKER_00: Would that move the needle?
40:16
SPEAKER_00: And actually,
40:17
that was how we developed this initial notion of a focused
40:20
research organization,
40:22
SPEAKER_00: which is in a nutshell,
40:23
it's just if you took a vertically integrated series A biotech
40:27
startup and a really
40:29
important nonprofit public benefit science goal and you put
40:32
them in a blender,
40:33
SPEAKER_00: what would you get out?
40:34
SPEAKER_00: You would get out something like E11.
40:37
SPEAKER_01: Yeah, which we could talk about froze.
40:40
SPEAKER_01: I'm going to borrow you to talk about brains,
40:43
but I am going to direct people to an episode that we did with
40:50
Adam Marlestone and
40:52
SPEAKER_01: Anastasia Gomic,
40:53
SPEAKER_01: who run Convergent Research,
40:56
SPEAKER_01: and it's episode 22,
40:57
and you can learn all about focused research organizations and
41:02
sort of the
41:02
SPEAKER_01: reasoning behind this.
41:03
SPEAKER_01: Yeah.
41:04
SPEAKER_01: But yeah,
41:05
SPEAKER_01: so yeah,
41:05
SPEAKER_01: I mean,
41:05
it fits into this category of somebody maybe too expensive for
41:12
a university to go
41:14
SPEAKER_01: after,
41:15
might not have totally clear kind of like product on the other
41:20
side for a venture
41:21
SPEAKER_01: capitalist to definitely fund,
41:23
SPEAKER_01: but is extremely important to do and somebody should put money behind it.
41:30
SPEAKER_00: Yeah.
41:31
SPEAKER_00: I mean,
41:33
SPEAKER_00: to put a little bit of a spin on it,
41:36
SPEAKER_00: my spin on it,
41:36
SPEAKER_00: it's like the project is pre-commercial.
41:38
SPEAKER_00: It doesn't make sense to do as a startup yet,
41:40
but it does require more coordination and more vertical
41:43
integration than you can
41:44
SPEAKER_00: typically find in academia.
41:45
SPEAKER_00: And we're not the first people to do this, right?
41:47
Like the Allen Institute has a long track record of doing team
41:51
science projects.
41:52
SPEAKER_00: Janelia, The Broad, they all have team science projects.
41:54
But I think the one thing that we really noticed is that
41:57
startup culture is a
41:59
SPEAKER_00: little bit different than,
42:00
SPEAKER_00: you know,
42:00
SPEAKER_00: nonprofit academic institute culture.
42:03
SPEAKER_00: And,
42:03
SPEAKER_00: you know,
42:03
SPEAKER_00: the FROs are typically,
42:05
SPEAKER_00: you know,
42:05
SPEAKER_00: focused research organizations are typically time bound.
42:07
SPEAKER_00: And so that creates a real sense of urgency to get the job done.
42:10
SPEAKER_00: Like five years-ish and yeah.
42:12
SPEAKER_00: And there's this pressure to, uh,
42:14
SPEAKER_01: keep it small,
42:14
SPEAKER_01: tight,
42:15
SPEAKER_01: focused,
42:16
SPEAKER_01: hardworking,
42:17
SPEAKER_01: um,
42:18
SPEAKER_01: some of that like startup buzz to charge after this goal.
42:21
SPEAKER_01: And then,
42:22
SPEAKER_01: and then you mentioned,
42:22
SPEAKER_01: I mean,
42:22
SPEAKER_01: other people have done it,
42:23
SPEAKER_01: but then some folks like Eric Schmidt and James Fickle,
42:28
SPEAKER_01: um,
42:28
SPEAKER_01: come to mind as,
42:29
SPEAKER_01: as people who are willing to put money behind a few of these.
42:33
SPEAKER_00: Yeah.
42:33
So,
42:33
so 11 bio was anchored by Eric and Wendy Schmidt,
42:37
um,
42:37
very generously.
42:38
And,
42:39
uh,
42:39
you know,
42:39
since then,
42:40
you know,
42:40
uh,
42:40
many others have,
42:41
you know,
42:42
uh,
42:43
SPEAKER_00: understood what we're doing and have wanted to support us.
42:45
And so James Fickle and Protocol Labs and the Bechtel
42:49
Foundation and the Hearst
42:50
Foundations and others are now very interested in the project
42:55
that we are pursuing,
42:56
which is to really try to solve some of these fundamental
42:59
problems around how do we
43:01
SPEAKER_00: make neurons different colours,
43:03
SPEAKER_00: enough different colours that you can actually trace them and
43:07
SPEAKER_00: then apply that notion to make the cost of connectomics much cheaper.
43:11
SPEAKER_00: And then there's one other thing that is very exciting,
43:13
SPEAKER_00: which is when you're using light microscopes,
43:16
you can use these different colors to identify different
43:19
neurons and distinguish
43:20
SPEAKER_00: them from each other.
43:20
But you can also use different wavelengths and different colors
43:24
to look at other
43:24
SPEAKER_00: things in the sample.
43:25
SPEAKER_00: And so for the first time,
43:26
SPEAKER_00: we can actually bring molecular information,
43:28
SPEAKER_00: biomarkers or synaptic markers or things that tell you cell type,
43:33
SPEAKER_00: into connectomics assays.
43:35
SPEAKER_00: And so that's like a new capability that hasn't existed before now.
43:39
SPEAKER_01: So if there's people who are keeping up,
43:41
SPEAKER_01: I'm sure they're scratching their head a little bit,
43:44
SPEAKER_01: which is that you,
43:46
SPEAKER_01: you said,
43:46
SPEAKER_01: where you were trying to do a hundred million neurons in a mouse brain,
43:51
you were running out of colors and these other techniques after
43:55
like a hundred,
43:55
SPEAKER_01: I guess.
43:56
Um,
43:57
so how on earth do you possibly even begin to mark up a hundred
44:03
million things differently?
44:05
Yeah,
44:06
so for people who are interested,
44:08
we just dropped our first preprint.
44:10
SPEAKER_00: So E11, you can
44:14
SPEAKER_00: find a link on our website or you can check it out on bioarchive.
44:16
SPEAKER_00: Yeah, yeah, I'll put it up too.
44:18
SPEAKER_00: Yep.
44:18
Describes,
44:19
you know,
44:19
the methods that we're using in order to make this work.
44:22
SPEAKER_00: But in short,
44:23
SPEAKER_00: you know,
44:23
we've extended the number of possible colors by splitting up
44:30
the assay into
44:31
SPEAKER_00: sequential cycles.
44:32
And this is the same idea that,
44:34
you know,
44:35
how DNA sequencers work actually in many ways.
44:38
So when you're sequencing DNA,
44:40
you know,
44:40
ideally you would want like one color per
44:43
SPEAKER_00: base or something like that,
44:44
SPEAKER_00: but there's no way you can split up the visible spectrum in that way.
44:47
SPEAKER_00: And so what you do is you have four colors,
44:50
SPEAKER_00: A,
44:50
SPEAKER_00: C,
44:50
SPEAKER_00: T or G,
44:51
and you look at the next base in the chain and it'll light up
44:55
as one of those four
44:55
SPEAKER_00: colors.
44:56
SPEAKER_00: And then you'll extinguish that.
44:59
SPEAKER_00: And then you'll take another step down the chain and you'll look,
45:01
SPEAKER_00: is it red,
45:01
SPEAKER_00: green,
45:02
SPEAKER_00: blue,
45:02
SPEAKER_00: or yellow?
45:03
SPEAKER_00: And you'll do that sequentially.
45:05
SPEAKER_00: And we're doing something similar.
45:06
So we have these different proteins that we can deliver to the
45:10
brain that are like
45:11
SPEAKER_00: fluorescent proteins that can be read out in sets,
45:15
SPEAKER_00: say five at a time or four at a time.
45:17
And so then we can,
45:18
for the first set of those proteins,
45:20
say,
45:21
is it red,
45:21
green,
45:22
blue,
45:22
or yellow?
45:23
SPEAKER_00: And then we can extinguish that and then we can read out the second set.
45:26
SPEAKER_00: And so we can do that in a sequential way.
45:29
And each time you do this,
45:31
you exponentially increase the number of possible combinations.
45:35
SPEAKER_00: And so it's the difference between colors and color codes.
45:39
SPEAKER_00: And so we're reading out colored codes.
45:41
SPEAKER_00: And what we've found is that we can make at this point in time,
45:44
18 of these different combination or like building blocks that
45:49
can be combined in
45:51
SPEAKER_00: combinatorial ways.
45:52
SPEAKER_00: So if there's 18 different proteins,
45:53
SPEAKER_00: there's two to the 18 different possible combinations,
45:56
SPEAKER_00: which is,
45:57
SPEAKER_00: you know,
45:57
SPEAKER_00: 260 something thousand combinations.
46:00
SPEAKER_00: So this is,
46:00
SPEAKER_00: you know,
46:02
like hundreds of times or even thousands of times more than
46:05
what people have been
46:06
SPEAKER_00: able to do previously.
46:07
SPEAKER_01: And how do you,
46:08
SPEAKER_01: when you talk about the barcoding,
46:10
SPEAKER_01: the tagging,
46:11
SPEAKER_01: and putting these proteins in the brain,
46:14
SPEAKER_01: if I remember right,
46:16
SPEAKER_01: you're using a virus more or less,
46:19
SPEAKER_01: right?
46:19
SPEAKER_01: I mean,
46:19
SPEAKER_01: you're injecting brain tissue with the virus,
46:22
SPEAKER_01: which is carrying the protein to an individual neuron.
46:27
SPEAKER_00: Is that the right way to think about it?
46:28
SPEAKER_00: Yeah,
46:29
we as a field have gotten remarkably good at using viruses to
46:33
deliver genes to the
46:34
SPEAKER_00: mouse brain.
46:35
SPEAKER_00: And these are called adeno-associated viruses, AAVs.
46:38
SPEAKER_00: They're very safe.
46:40
SPEAKER_00: They're just a workhorse in biology these days.
46:42
SPEAKER_00: People use them to deliver genes to the brain.
46:45
They don't replicate,
46:46
they just sort of hang out after they deliver the gene of
46:49
interest.
46:50
SPEAKER_00: And there has been a lot of work to
46:55
SPEAKER_00: develop these viruses so they can be administered systemically,
46:58
SPEAKER_00: so they can cross the blood-brain barrier,
47:00
SPEAKER_00: so they can get into many different neurons in the brain.
47:03
And so this is really a frontier in how we can genetically
47:09
engineer organisms.
47:11
And so,
47:11
okay,
47:12
so you take,
47:14
you've got a mouse brain,
47:16
you put in the virus.
47:18
SPEAKER_00: Are you putting in many viruses or one?
47:20
SPEAKER_00: It's a pool of viruses.
47:21
SPEAKER_00: It's a pool of viruses.
47:23
I mentioned there's 18 different variants in this publication
47:27
that you can check out.
47:28
SPEAKER_00: And you make a pool where each of them is present at an equal ratio.
47:35
SPEAKER_00: And then when you deliver them to the brain,
47:37
SPEAKER_00: The barcodes are delivered essentially randomly.
47:40
Each cell will just get some subset of that total set of
47:45
different viruses.
47:45
SPEAKER_00: So, you know, cell one might get, you know, variants one, two, and four.
47:49
SPEAKER_00: And,
47:50
SPEAKER_00: you know,
47:50
SPEAKER_00: now the barcode is one,
47:51
SPEAKER_00: zero,
47:52
SPEAKER_00: one,
47:52
SPEAKER_00: one,
47:53
SPEAKER_00: zero,
47:53
SPEAKER_00: zero,
47:54
SPEAKER_00: zero,
47:54
SPEAKER_00: all the way up to 18 and so on.
47:56
SPEAKER_00: And so you basically use like random,
47:59
SPEAKER_00: called stochastic co-infection,
48:01
where it just means that each cell gets infected multiple times
48:04
to deliver the
48:04
SPEAKER_00: barcode randomly.
48:05
And so some neuron is absorbing a different combination of the
48:15
proteins from its neighbor.
48:18
SPEAKER_01: I mean,
48:18
this is what's helping you distinguish because each neuron is
48:22
going to pick up
48:23
SPEAKER_01: something different in this process.
48:25
SPEAKER_01: And so then because they fluoresce and give off this different
48:31
SPEAKER_01: color,
48:31
instead of getting this black and white map where everything
48:34
looks the same,
48:35
now you're starting to see differentiation between the
48:39
different neurons in the
48:40
SPEAKER_01: brain.
48:40
SPEAKER_01: And you can follow a blue wire and a green wire, essentially.
48:48
SPEAKER_00: Yeah.
48:48
SPEAKER_00: And so we apply this.
48:49
SPEAKER_00: This is our first pilot study.
48:51
SPEAKER_00: We apply it in mouse hippocampus,
48:53
SPEAKER_00: which is,
48:54
SPEAKER_00: you know,
48:54
SPEAKER_00: the hippocampus is very important for learning and for memory.
48:57
SPEAKER_00: And I should say that this is a very collaborative study.
49:00
SPEAKER_00: Actually, we call it PRISM.
49:01
One of the most fun things about developing PRISM was that it
49:05
was this long-term
49:07
SPEAKER_00: collaboration that started in Ed Boyden's lab,
49:11
SPEAKER_00: Sam Rodriguez,
49:12
SPEAKER_00: Juergen Kornfeld,
49:13
SPEAKER_00: who's now at Max Planck,
49:14
SPEAKER_00: Sam went to the Crick and is now at Future House in San Francisco,
49:18
SPEAKER_00: and Boban in Ed's lab.
49:20
SPEAKER_00: and I all work together on the initial instantiation of this.
49:23
SPEAKER_00: And then many others have brought their intellectual lineage to it.
49:28
So Kathleen Lieper and Johann Winnebst and Sven Schrockenbrot
49:31
and Arlo Sheridan are
49:33
all new members of B11 who have brought different aspects of
49:37
barcoding or expansion
49:38
SPEAKER_00: microscopy or machine learning to this problem.
49:41
But we all got together to perform this initial pilot study
49:45
where we delivered
49:46
SPEAKER_00: these barcodes to the mouse hippocampus and then
49:51
applied and optimized the expansion microscopy protocols in
49:55
combination with these
49:56
SPEAKER_00: barcodes,
49:56
because you have to have both of them working together in order
49:59
to actually be able
50:00
to look at these neurons in a way that accommodates the fact
50:05
that light is big and
50:06
SPEAKER_00: neurons are small.
50:07
SPEAKER_00: So that required a lot of co-optimization.
50:11
SPEAKER_00: And then we needed new machine learning methods and AI methods
50:14
that would actually take this multicolor information and use it
50:17
automatically to
50:19
SPEAKER_00: correct errors.
50:20
SPEAKER_00: And we can do this in two different ways.
50:22
SPEAKER_00: One is when you're actually drawing the boundaries around the shapes and
50:26
SPEAKER_00: identifying the boundaries.
50:27
SPEAKER_00: And then, you know,
50:29
I mentioned earlier that when you are correcting the outputs of
50:34
normal computer
50:35
SPEAKER_00: vision models for connectomics,
50:36
SPEAKER_00: you have humans who are proofreading the outputs and correcting them.
50:39
SPEAKER_00: And so we actually can use this color information again in an automatic
50:42
proofreading step where there's a model or an algorithm that is
50:47
able to identify
50:49
when errors have occurred and then use the barcodes to correct
50:52
those errors.
50:54
SPEAKER_01: Okay, so many questions.
50:55
SPEAKER_01: Okay,
50:55
SPEAKER_01: so you mentioned,
50:57
SPEAKER_01: okay,
50:58
SPEAKER_01: in this paper,
51:02
SPEAKER_01: you're talking about part of a mouse brain,
51:04
SPEAKER_01: you're talking about the levels you've been able to get to 2 to the 18th,
51:08
SPEAKER_01: I think.
51:09
SPEAKER_01: Is that right?
51:10
SPEAKER_01: Yeah, we can encode 2 to the 18th combination.
51:12
SPEAKER_01: Right, okay.
51:13
And then if you wanted to do an entire mouse brain,
51:17
the numbers have to go up,
51:19
right?
51:19
SPEAKER_01: Yeah.
51:20
SPEAKER_00: So this is a great question.
51:22
SPEAKER_00: So there's a couple of different levers.
51:26
SPEAKER_00: So one is to increase the number of possible combinations.
51:29
SPEAKER_00: And with each new protein variant that we discover,
51:31
SPEAKER_00: and we're running a pipeline at E11 to discover these variants,
51:34
SPEAKER_00: with each new variant,
51:36
SPEAKER_00: we double the number of possible barcodes.
51:38
SPEAKER_00: So this grows very exponentially.
51:39
SPEAKER_00: So with 18, you have 262,000.
51:42
SPEAKER_00: With 30, you have about a billion.
51:44
SPEAKER_00: We've been finding these at a rate of one per couple months.
51:48
SPEAKER_00: So this has the potential to grow quite quickly.
51:51
SPEAKER_00: But I think there's another question, which is how many do you need?
51:54
SPEAKER_00: Because we're not throwing out the shape information as well.
51:57
SPEAKER_00: We're still using the shapes of the neurons and other methods from...
52:03
electron microscopy,
52:04
connectomics,
52:04
computer vision in combination with these barcodes.
52:07
SPEAKER_00: And so you can still trace the shapes as well.
52:09
SPEAKER_00: So I think how many barcodes do you need?
52:12
How much shape information do you need is like a research
52:14
problem that we're pursuing right now.
52:16
SPEAKER_00: But what we have found is that in this particular pilot study,
52:21
SPEAKER_00: when you apply these barcodes,
52:23
we improve the accuracy of tracing the neurons by almost an
52:27
order of magnitude.
52:29
And we like see a clear path to extending that further to that
52:33
100 X that I'm talking about.
52:34
So,
52:35
you know,
52:35
we're,
52:35
we're,
52:36
you know,
52:36
one log there,
52:37
we need to get a second log there.
52:39
SPEAKER_00: And when we, when we do and,
52:43
SPEAKER_00: when we combine that with other methods that are also,
52:45
SPEAKER_00: you know,
52:46
SPEAKER_00: being developed right now for EM,
52:47
SPEAKER_00: for electron microscopy,
52:49
SPEAKER_00: I think,
52:49
SPEAKER_00: you know,
52:49
SPEAKER_00: we'll be really making a dent in this,
52:52
SPEAKER_00: you know,
52:52
SPEAKER_00: big bottleneck,
52:53
SPEAKER_00: which is the cost of tracing neurons.
52:55
SPEAKER_01: Yeah.
52:55
SPEAKER_01: So,
52:55
SPEAKER_01: I mean,
52:56
SPEAKER_01: there's,
52:57
SPEAKER_01: as far as I understand it,
52:58
SPEAKER_01: I mean,
52:58
SPEAKER_01: and you mentioned this one study that,
53:01
SPEAKER_01: it's just people are like,
53:02
SPEAKER_01: oh,
53:02
SPEAKER_01: well,
53:03
SPEAKER_01: we're not getting to a human connectome,
53:06
SPEAKER_01: like,
53:06
anytime soon unless somebody figures something out or has
53:10
infinite money what what
53:12
is like the best um what's the most public quite you know date
53:19
or or goal for the
53:20
SPEAKER_01: the mouse brain a whole mouse brain at the moment i i think that's a
53:25
interesting topic of discussion that different people in the
53:30
field have different
53:31
SPEAKER_00: reference points for.
53:31
I think most people would be surprised if it showed up sooner
53:35
than five years from
53:36
SPEAKER_00: now,
53:37
SPEAKER_00: and also surprised if it took more than 15.
53:39
SPEAKER_00: But I think
53:42
People were also very surprised in genomics when the cost of
53:46
sequencing fell 10,000
53:47
SPEAKER_00: fold in five years.
53:48
SPEAKER_00: And so when new technologies come along,
53:50
it's sometimes difficult to predict exactly how quickly they'll
53:54
be able to reshape
53:55
SPEAKER_00: the landscape.
53:56
SPEAKER_01: And then,
53:57
SPEAKER_01: granted,
53:58
your company's dedicated to this and you've dedicated this
54:01
portion of your life to
54:02
SPEAKER_01: this,
54:02
SPEAKER_01: so possibly some bias.
54:04
SPEAKER_01: But,
54:05
SPEAKER_01: you know,
54:05
given what you've seen with this research and the breakthroughs
54:09
that you guys have
54:10
SPEAKER_01: had,
54:10
SPEAKER_01: how has that altered your thinking?
54:13
SPEAKER_01: What is your,
54:13
SPEAKER_01: like,
54:14
most optimistic case for your technology being able to do a
54:19
whole mouse brain?
54:20
SPEAKER_01: Or do we still not know if that's feasible?
54:24
SPEAKER_00: I think there's a lot of uncertainty here.
54:27
And part of the fun of running a focused research organization
54:31
is we get to test
54:32
SPEAKER_00: these hypotheses very empirically.
54:34
SPEAKER_00: We get to go and find out how good is the technology going to be.
54:39
And so I would say a lot of the hard problem has been just
54:44
figuring out how to do this at all.
54:46
SPEAKER_00: And I would be less...
54:51
SPEAKER_00: I would say that five to 15 years seems right for any technology,
54:56
SPEAKER_00: but if we can make that happen in more like five years rather than in 15,
55:00
that would be really transformative for all of the people who
55:03
would benefit from
55:04
SPEAKER_00: having connectomes sooner.
55:06
SPEAKER_00: And there's another thing.
55:07
SPEAKER_00: We've been talking a lot about what would it take to do an entire brain.
55:14
The medical value of being able to do parts of the brain or
55:19
many sub-volumes of
55:20
SPEAKER_00: brain is really important.
55:22
SPEAKER_00: For example,
55:23
people are working on cell therapies where in cases of
55:27
Parkinson's or traumatic
55:28
SPEAKER_00: brain injury or stroke,
55:30
you need to be able to put stem cells into the brain that then
55:34
integrate and regrow
55:35
SPEAKER_00: the
55:35
and connect back up in many of the same ways that you would
55:40
want the healthy cells.
55:42
And the nice thing about the cell therapies is you have a lot
55:44
of control over how
55:46
SPEAKER_00: you engineer those cells.
55:48
SPEAKER_00: You can actually...
55:49
conceive of putting things like barcodes into those neurons or
55:52
into those stem
55:53
SPEAKER_00: cells and then deliver them to the brain.
55:56
SPEAKER_00: And so there's a lot to do when you have multiple colors.
56:02
You can see how those cell therapies are integrating and
56:06
connecting and rewiring
56:08
SPEAKER_00: using that color information.
56:10
SPEAKER_00: You can use that to distinguish it from the host tissue.
56:12
SPEAKER_00: And so this could be a
56:18
a medical application where it could happen a lot sooner than
56:22
being able to map the
56:23
SPEAKER_00: entire brain.
56:23
And so I think we should think about different ways that
56:26
connectomics technology
56:27
SPEAKER_00: can be a little unbundled from the main push,
56:31
SPEAKER_00: which is to get the whole brain.
56:32
SPEAKER_00: Don't cop out on me, Andrew.
56:33
SPEAKER_00: I want the whole human brain.
56:35
SPEAKER_00: Oh, I mean, the mission of the company is to make that happen.
56:39
SPEAKER_01: All right,
56:40
SPEAKER_01: I'm just,
56:40
SPEAKER_01: I'm looking up a message at the moment from Sebastian Sung,
56:45
SPEAKER_01: who worked,
56:47
SPEAKER_01: I think it's fair to say he was like a central figure on the fly.
56:50
SPEAKER_01: He wrote the book Connectome.
56:51
SPEAKER_01: He wrote the book Connectome.
56:52
SPEAKER_01: It's a great book.
56:53
SPEAKER_01: But the fly project came after the book and was a pretty central figure.
57:00
SPEAKER_01: I don't want to get this wrong.
57:03
SPEAKER_01: Yeah.
57:05
So I'm just looking it up because if I remember correctly and I
57:08
will launch my
57:09
SPEAKER_01: question and then I'll like dig in and see if I'm right or not.
57:13
SPEAKER_01: But I think coming out of the fly,
57:15
his argument to me was that the AI software had been
57:21
surprisingly good and
57:23
incredibly helpful and that he thought we get over a lot of
57:28
these costs and
57:29
SPEAKER_01: technical issues just through the study process.
57:33
um sometimes exponential improvement in in ai software yeah i i
57:39
would say you know
57:40
um it was surprising in many ways how little human proofreading
57:45
was needed to get
57:46
you know that fly brain only 33 person years versus you know
57:49
the the many thousands
57:50
SPEAKER_00: that you know would have been required before advances in ai
57:54
SPEAKER_00: So that sort of brings us to the other challenge,
57:57
SPEAKER_00: which is attacking the cost of tracing is the biggest bottleneck.
58:01
SPEAKER_00: And we need just many different methods.
58:03
SPEAKER_00: We need to be able to combine improvements in AI,
58:07
SPEAKER_00: like the kind that Sebastian was talking about,
58:09
SPEAKER_00: with other modalities.
58:11
SPEAKER_00: Like when we're literally injecting more information into the brain,
58:13
that will just make it easier and more effective because we
58:17
want to do many brains,
58:18
SPEAKER_00: right?
58:18
SPEAKER_00: So if we can
58:19
Like,
58:19
why would you say no to another factor of 10 reduction in cost,
58:24
for example?
58:24
SPEAKER_00: But the other really challenging thing,
58:26
SPEAKER_00: which we haven't touched on,
58:27
is that slicing the brain into these thin sections or ablating
58:31
it is actually a
58:32
very precision engineering process that is challenging to do in
58:39
a scalable way
58:41
SPEAKER_00: thousands and thousands of times in a row.
58:44
SPEAKER_00: And the Microns project,
58:45
SPEAKER_00: which was this cubic millimeter of mouse brain tissue,
58:50
SPEAKER_00: they took a sub volume of the mouse brain,
58:51
SPEAKER_00: a cubic millimeter,
58:52
SPEAKER_00: and then sliced it into 28,000 individual thin sections.
58:57
SPEAKER_00: And this was like an incredible engineering achievement.
59:00
SPEAKER_00: And also the diamond knife jammed about 40% of the way through.
59:04
SPEAKER_00: And so what we actually have is a sub volume,
59:07
SPEAKER_00: which is like 400 microns thick and a sub volume,
59:10
SPEAKER_00: which is like 600 microns thick.
59:12
SPEAKER_00: And we don't actually have
59:13
SPEAKER_00: the wires across that gap of six slices in the middle.
59:17
SPEAKER_00: You just can't trace it if there's nothing there.
59:20
And so I think one of the additional challenges with this
59:25
approach to slicing is
59:27
SPEAKER_00: that if you're gonna do one mouse brain,
59:30
you need to do that 28,000 slice problem 500 times for all of
59:35
the one cubic
59:36
SPEAKER_00: millimeter sub volumes of the mouse brain.
59:37
SPEAKER_00: And the chances of making an error are still high.
59:41
And so one of the...
59:42
Yeah,
59:43
well,
59:43
because these aren't machines you just like go buy,
59:45
right?
59:45
SPEAKER_01: I mean, they're sort of like hand built more or less.
59:48
SPEAKER_00: Yeah, I mean, it took months to do the slicing.
59:50
SPEAKER_00: Yeah.
59:51
So one of the interesting things about this expansion
59:55
microscopy approach is that
59:57
SPEAKER_00: you can work with much thicker sections.
1:00:00
Instead of working with 50 nanometer sections,
1:00:02
you can work with 50 micron sections.
1:00:03
SPEAKER_00: So the sections are a thousand times bigger.
1:00:06
SPEAKER_00: And so now when you're talking about a cubic millimeter,
1:00:08
SPEAKER_00: you're talking about maybe 20 sections rather than 28,000.
1:00:12
SPEAKER_00: So,
1:00:12
SPEAKER_00: you know,
1:00:12
that's the sort of difference where I think new technology is
1:00:18
still going to be
1:00:19
SPEAKER_00: really helpful.
1:00:20
SPEAKER_00: You know,
1:00:20
SPEAKER_00: because we're asking this question,
1:00:23
SPEAKER_00: not necessarily like,
1:00:25
SPEAKER_00: what is the method that will get the first connectome,
1:00:28
SPEAKER_00: but like,
1:00:29
SPEAKER_00: what is the technology that will give us that next generation sequencing,
1:00:32
SPEAKER_00: like bend the cost curve moment that will make it so cheap that,
1:00:35
SPEAKER_00: you know,
1:00:36
SPEAKER_00: anyone can do it.
1:00:37
And I think,
1:00:37
you know,
1:00:38
that's one of the really interesting questions in connectomics.
1:00:40
SPEAKER_01: Okay.
1:00:42
SPEAKER_01: Will you get off the hook a little bit here?
1:00:45
SPEAKER_01: My messages with Sebastian from,
1:00:47
SPEAKER_01: this is from months ago,
1:00:49
SPEAKER_01: I think when the fly came out,
1:00:51
SPEAKER_01: he says,
1:00:51
I have a somewhat different take regarding the value
1:00:55
proposition of E11 and its
1:00:56
SPEAKER_01: competitors,
1:00:58
SPEAKER_01: but I've got to run now and we'll answer your questions later.
1:01:02
SPEAKER_01: And he didn't.
1:01:04
SPEAKER_01: Do you have competitors?
1:01:06
SPEAKER_01: He lists Penluminate.
1:01:11
SPEAKER_01: Oh, I love Penluminate.
1:01:12
SPEAKER_01: Okay.
1:01:13
SPEAKER_01: Well, I mean, he just sort of says they're impressive.
1:01:15
SPEAKER_01: I don't know if competitor is fully fair.
1:01:18
SPEAKER_01: And then he links me to a paper by a researcher named Tavacoli.
1:01:25
SPEAKER_01: Yeah.
1:01:25
SPEAKER_01: Okay.
1:01:26
SPEAKER_01: Yeah.
1:01:26
SPEAKER_01: I don't know.
1:01:27
SPEAKER_01: Are there, are there different techniques?
1:01:28
SPEAKER_00: So like I said, you know, expansion microscopy started a field, right?
1:01:31
SPEAKER_00: So, you know, many, like it's been around for 10 years now.
1:01:35
And in the recent past,
1:01:38
there has been some pretty exciting advances where you can make
1:01:43
these expansion microscopy samples look more and more like
1:01:47
electron microscopy.
1:01:48
SPEAKER_00: So you can,
1:01:48
SPEAKER_00: you know,
1:01:49
SPEAKER_00: you can expand it more,
1:01:50
SPEAKER_00: you know,
1:01:51
SPEAKER_00: in our prism study,
1:01:51
SPEAKER_00: which we just released,
1:01:52
SPEAKER_00: you know,
1:01:53
SPEAKER_00: we expand it five fold.
1:01:55
SPEAKER_00: If you expand it,
1:01:56
SPEAKER_00: you know,
1:01:57
SPEAKER_00: 16 or 20 fold,
1:01:58
SPEAKER_00: you can actually see the,
1:02:00
SPEAKER_00: you know,
1:02:01
SPEAKER_00: it looks like EM essentially.
1:02:03
SPEAKER_00: And so, you know, this was pioneered by a couple of people, um, uh,
1:02:09
SPEAKER_00: Ans M.
1:02:10
SPEAKER_00: Saad,
1:02:10
SPEAKER_00: who founded Panluminate,
1:02:12
SPEAKER_00: developed this pan-ext,
1:02:15
SPEAKER_00: I think that's what it's called,
1:02:17
SPEAKER_00: pan stain.
1:02:18
SPEAKER_00: It's called pan because it stains all of the proteins,
1:02:21
SPEAKER_00: which stains the cell morphology in this way,
1:02:23
SPEAKER_00: and it looks like EM.
1:02:25
SPEAKER_00: Hugo Damstra at E11 did something similar with T-Rex.
1:02:28
SPEAKER_00: It's another expansion technique.
1:02:31
SPEAKER_00: I'm getting into the weeds here.
1:02:34
SPEAKER_00: One of the most interesting things was out of IST Austria,
1:02:37
SPEAKER_00: Moe Tacavoli last year developed this technique called Likon,
1:02:41
SPEAKER_00: which took it a step further.
1:02:43
SPEAKER_00: And he worked with Varen Jane's group, and this was in the Donzel lab.
1:02:47
SPEAKER_00: So Moe Tacavoli and Johan Donzel worked with Varen Jane at Google.
1:02:50
to apply the EM tools to these morphology stains in order to
1:02:55
actually get the connectome out.
1:02:57
SPEAKER_00: And actually, that was one of the missing pieces.
1:02:59
Before that,
1:03:00
people hadn't managed to actually get connectomes out of
1:03:03
expansion microscopy.
1:03:05
SPEAKER_00: So that's actually one of the why now moments.
1:03:06
SPEAKER_00: And this is one of the things that we have integrated into our pipeline.
1:03:11
SPEAKER_00: And our thesis is that all of the problems,
1:03:14
the challenges that EM faces with tracing these neurons through
1:03:17
black and white
1:03:18
SPEAKER_00: images
1:03:19
SPEAKER_00: are also challenges that, you know, these pan stains have.
1:03:22
And,
1:03:23
you know,
1:03:23
you're going to have similar error rates and similar challenges
1:03:25
tracing neurons.
1:03:27
And so if you can bring in this additional color information,
1:03:29
it's only going to help.
1:03:30
SPEAKER_00: And so, you know, it's like a yes and.
1:03:33
SPEAKER_00: Got it.
1:03:34
SPEAKER_01: There's something...
1:03:35
SPEAKER_01: Okay,
1:03:36
SPEAKER_01: well,
1:03:36
SPEAKER_01: I feel like we've...
1:03:37
SPEAKER_01: Sebastian ordered me to,
1:03:40
SPEAKER_01: you know,
1:03:40
at least talk about some of these other folks working on some
1:03:45
parts of this.
1:03:45
SPEAKER_01: So we've done that.
1:03:48
SPEAKER_01: When I was at
1:03:49
SPEAKER_01: I don't know if I want to get Jeff Lickman in trouble or not.
1:03:52
SPEAKER_01: I don't think I will.
1:03:53
SPEAKER_01: I mean,
1:03:53
I found a strange part of this was that the amount of compute
1:03:57
is so high that
1:03:59
Harvard was like basically in line to get space on some Google
1:04:05
cluster somewhere to
1:04:06
SPEAKER_01: do all the...
1:04:07
SPEAKER_01: calculations that they needed in the imaging,
1:04:10
SPEAKER_01: you know,
1:04:10
SPEAKER_01: to,
1:04:10
SPEAKER_01: to stitch all these,
1:04:11
SPEAKER_01: these things together.
1:04:12
SPEAKER_01: And,
1:04:13
SPEAKER_01: um,
1:04:14
SPEAKER_01: yeah,
1:04:14
I guess I was sort of surprised that an institution of even
1:04:18
Harvard had to,
1:04:19
SPEAKER_01: yeah,
1:04:19
SPEAKER_01: I was like,
1:04:20
SPEAKER_01: well,
1:04:20
SPEAKER_01: how do you,
1:04:20
SPEAKER_01: how do you move up and down in the line?
1:04:22
SPEAKER_01: I was like, well, you gotta kind of like call in favors.
1:04:25
SPEAKER_01: You gotta know somebody.
1:04:26
SPEAKER_01: And, and, um, I found it interesting that, um,
1:04:29
This is such a big computing problem that you're like horse
1:04:32
trading to get space on
1:04:34
SPEAKER_01: these supercomputers to do this work.
1:04:38
And that also seemed like a massive bottleneck to me and just a
1:04:43
strange situation in some ways.
1:04:45
SPEAKER_00: I mean,
1:04:45
I think it's not that strange in the sense that when you are
1:04:48
pioneering and trying
1:04:51
SPEAKER_00: a lot of the stuff for the first time,
1:04:53
SPEAKER_00: you haven't had the time to optimize the methods.
1:04:57
SPEAKER_00: And so,
1:04:58
SPEAKER_00: you know,
1:04:58
a lot of the more recent approaches to machine learning for
1:05:04
neuron tracing were
1:05:05
SPEAKER_00: pioneered by,
1:05:06
SPEAKER_00: you know,
1:05:07
people who had access to Google resources because just getting
1:05:10
it to work at all is
1:05:11
SPEAKER_00: step one and then getting it to work cheaply is steps two through end.
1:05:16
And folks like Arla Sheridan at E11 and Jan Funke at Janelia
1:05:22
have been working on
1:05:24
SPEAKER_00: different versions that maybe aren't quite as accurate but are much more
1:05:28
SPEAKER_00: computationally efficient.
1:05:29
SPEAKER_00: And I think it's just in general,
1:05:31
SPEAKER_00: you know,
1:05:31
SPEAKER_00: there's this like give and take where,
1:05:34
SPEAKER_00: you know,
1:05:34
people will come up with a new method that like breaks new
1:05:37
ground and then you
1:05:38
SPEAKER_00: have,
1:05:38
SPEAKER_00: you know,
1:05:39
additional work to optimize it and make it cheaper and make it
1:05:42
more effective.
1:05:43
SPEAKER_00: And so,
1:05:43
SPEAKER_00: you know,
1:05:44
SPEAKER_00: in that sense,
1:05:45
SPEAKER_00: I think it's not too surprising,
1:05:46
SPEAKER_00: but it just kind of underscores this point that,
1:05:48
SPEAKER_00: you know,
1:05:49
SPEAKER_00: we need to make Connectomics much cheaper and like develop new tools and,
1:05:54
SPEAKER_00: you know,
1:05:55
SPEAKER_00: orient our research towards making it really cost effective.
1:05:58
SPEAKER_01: Yeah.
1:05:58
SPEAKER_01: Okay.
1:05:59
SPEAKER_01: All right.
1:05:59
SPEAKER_01: For this last section, let's go philosophical for a second.
1:06:03
SPEAKER_01: I just want to ask you my big picture questions.
1:06:08
Sometimes I run across people,
1:06:10
like I talked about on the AI front,
1:06:13
that think if we...
1:06:15
The more we know about the human brain,
1:06:17
the better we can inform our AI models.
1:06:21
I always get the gist of it,
1:06:23
but I sometimes stumble upon the mechanics of how that works.
1:06:28
I know another group that is trying to stick a ton of probes in
1:06:34
a brain to create a digital...
1:06:37
SPEAKER_01: a very complete digital model of the brain.
1:06:39
SPEAKER_01: And again, the hope is some kind of alignment type thing.
1:06:43
SPEAKER_01: Yeah, I stumble when I, I mean, again, I get the gist, right?
1:06:47
SPEAKER_01: It's like, okay, if we
1:06:49
know more about what if we find consciousness in here somewhere
1:06:54
or if we know more
1:06:56
about what makes the human brain special and we can somehow
1:06:59
inject that into these
1:07:01
ai models that are very at least for me like although over time
1:07:07
i think i'm
1:07:07
changing my opinion but roughly based on the structure of a
1:07:10
human brain in the
1:07:11
SPEAKER_01: sense of just connecting lots of
1:07:14
Yeah,
1:07:15
but I saw you talking about the subject and you have thoughts
1:07:19
on this.
1:07:19
Can you explain where you might see this going and if you think
1:07:27
this knowledge does
1:07:28
SPEAKER_01: have some application in AI?
1:07:30
SPEAKER_01: Yeah.
1:07:31
I think there's a lot of excitement around this nascent field
1:07:36
of neuro AI,
1:07:38
SPEAKER_00: which is how can AI learn from neuroscience?
1:07:40
SPEAKER_00: How can neuroscience learn from AI?
1:07:42
We know from history that there has been an interplay between
1:07:47
neuroscience and AI.
1:07:48
SPEAKER_00: The original artificial neural networks that we use day in,
1:07:54
SPEAKER_00: day out these days were named because they resembled the brain.
1:07:59
SPEAKER_00: And,
1:07:59
SPEAKER_00: you know,
1:08:01
the convolutional neural networks that drove the deep learning
1:08:06
revolution were,
1:08:07
SPEAKER_00: you know,
1:08:07
very similar to or organized in very similar ways to the visual
1:08:12
cortex and the
1:08:12
SPEAKER_00: visual system.
1:08:13
SPEAKER_00: So there's definitely these parallels that exist.
1:08:17
SPEAKER_00: But it's a little hard to say that,
1:08:18
SPEAKER_00: you know,
1:08:19
SPEAKER_00: those advances in AI were caused by neuroscience or,
1:08:22
SPEAKER_00: you know,
1:08:23
SPEAKER_00: it directly led to the other.
1:08:25
SPEAKER_00: But it could have.
1:08:26
SPEAKER_00: There's definitely worlds in which that could have happened.
1:08:28
SPEAKER_00: And so I think one of the most interesting questions is,
1:08:31
SPEAKER_00: if you have a field which can very systematically map brain circuits,
1:08:37
SPEAKER_00: mine them,
1:08:37
SPEAKER_00: look for patterns and motifs and changes between different organisms,
1:08:43
SPEAKER_00: can you
1:08:45
drive cross-pollination to speed up the development of new or
1:08:50
safer AI tools that
1:08:52
SPEAKER_00: are based on the brain?
1:08:53
And can you actually do that in an intentional and systematic
1:08:57
way rather than in a
1:08:59
SPEAKER_00: serendipitous or loosely inspired way?
1:09:01
SPEAKER_00: And so I think that's a thesis that needs to be tested out.
1:09:06
But one of the challenges for a long time has been we just
1:09:09
haven't been able to have the data
1:09:11
SPEAKER_00: And so, you know, it's been hard to drive that cross-pollination.
1:09:14
SPEAKER_00: So what if you could do that in a systematic way?
1:09:16
SPEAKER_00: That's really interesting to me.
1:09:18
SPEAKER_01: Like, what have you made?
1:09:21
SPEAKER_01: Going back to that original thing,
1:09:22
SPEAKER_01: I mean,
1:09:22
SPEAKER_01: yes,
1:09:23
they called the original neural networks neural networks and
1:09:27
always talked about,
1:09:27
SPEAKER_01: you know,
1:09:28
SPEAKER_01: I don't know,
1:09:28
SPEAKER_01: for a long time,
1:09:29
SPEAKER_01: like I said,
1:09:29
SPEAKER_01: I found it unsatisfying because it was just like,
1:09:32
SPEAKER_01: ah.
1:09:32
SPEAKER_01: you just stitched a bunch of computing elements together and you,
1:09:35
SPEAKER_01: you know,
1:09:36
SPEAKER_01: it's like,
1:09:36
SPEAKER_01: okay,
1:09:37
SPEAKER_01: it's like mimics the brain,
1:09:38
SPEAKER_01: but in,
1:09:39
SPEAKER_01: in the most basic possible way.
1:09:41
SPEAKER_01: And then,
1:09:42
SPEAKER_01: and then as we moved along,
1:09:43
SPEAKER_01: I was like,
1:09:43
SPEAKER_01: oh God,
1:09:44
SPEAKER_01: well,
1:09:44
SPEAKER_01: I mean,
1:09:45
SPEAKER_01: apparently if you put enough complexity in this thing,
1:09:47
SPEAKER_01: it starts to do pretty interesting stuff that we can't quite explain.
1:09:51
SPEAKER_01: And it does seem to,
1:09:52
SPEAKER_01: um,
1:09:53
SPEAKER_01: whether you want to call it a parrot or not,
1:09:55
SPEAKER_01: I mean,
1:09:55
SPEAKER_01: it's,
1:09:56
SPEAKER_01: it's,
1:09:56
SPEAKER_01: it's clear.
1:09:56
SPEAKER_01: It's like thinking something on some level, you know?
1:10:00
SPEAKER_01: And, and so that makes me feel like, okay,
1:10:03
SPEAKER_01: This seems to be some kind of precursor or related to the brain.
1:10:08
SPEAKER_01: Yeah,
1:10:08
SPEAKER_01: I mean,
1:10:09
SPEAKER_01: what have you made of AI models over the last five years,
1:10:14
SPEAKER_01: just given your background in neuroscience?
1:10:17
And I don't know,
1:10:20
it's a giant question,
1:10:21
but I'm always curious how you read these things.
1:10:25
SPEAKER_00: Yeah, I mean, you know, they've been making phenomenal progress.
1:10:29
You know,
1:10:29
I think anybody who says they knew that it was coming would be,
1:10:35
you know,
1:10:36
either
1:10:38
SPEAKER_00: directly working on it or making it up after the fact.
1:10:41
I did not expect that the AI models would advance as quickly as
1:10:44
they did,
1:10:44
and I updated on that.
1:10:47
SPEAKER_00: I think there's a lot of things that they do well.
1:10:49
SPEAKER_00: There are some things that are still really hard, right?
1:10:51
SPEAKER_00: Continuous learning is not really a capability that these models have.
1:10:55
SPEAKER_00: They have to be trained, and then you have a snapshot.
1:11:00
SPEAKER_00: And so I think
1:11:02
there's still a lot to be learned from what biological systems
1:11:08
and living systems
1:11:09
SPEAKER_00: do that current models can't.
1:11:13
SPEAKER_00: And indeed, one of those things is we generally are
1:11:18
SPEAKER_00: pretty well aligned with each other as humans.
1:11:21
SPEAKER_00: We cooperate,
1:11:23
SPEAKER_00: we're generally prosocial,
1:11:24
SPEAKER_00: we generally are altruistic,
1:11:26
SPEAKER_00: and there exists other organisms which that is not true for.
1:11:33
And so what is special about the wiring of humans or other
1:11:37
prosocial animals
1:11:40
SPEAKER_00: that is different than antisocial animals.
1:11:43
SPEAKER_00: And if we can figure that out,
1:11:44
SPEAKER_00: if we can figure out what is special about those connectomes,
1:11:49
SPEAKER_00: that could,
1:11:51
in an intentional way rather than a serendipitous or loosely a
1:11:56
loose analogy way,
1:11:58
SPEAKER_00: inspire a new set of AI safety work.
1:12:02
Why do you,
1:12:02
you mentioned brain computer interfaces before and how exciting
1:12:05
they are.
1:12:06
SPEAKER_01: And obviously tons of money has been pumped in to whatever it is,
1:12:10
SPEAKER_01: 12 to 20 startups,
1:12:12
SPEAKER_01: all approaching BCIs in different ways.
1:12:15
Why do you choose to work on the problem you're working on as
1:12:19
opposed to getting
1:12:20
into that field where you can be quite directly interfacing
1:12:25
with and manipulating
1:12:27
SPEAKER_01: and learning from brains?
1:12:29
SPEAKER_00: Well,
1:12:29
SPEAKER_00: like I said,
1:12:29
I think there is a sense of urgency here around our ability to
1:12:36
interface with the
1:12:37
SPEAKER_00: brain is outstripping our ability to know what we're interfacing with.
1:12:41
SPEAKER_00: And so there's a lot of opportunity cost,
1:12:45
SPEAKER_00: I think,
1:12:46
in not having the maps of the brain that would be useful to
1:12:50
inform where we should
1:12:52
SPEAKER_00: be
1:12:53
plugging in our BCIs or how we should be stimulating particular
1:12:57
circuits to restore function.
1:12:59
SPEAKER_00: And this is kind of a public good.
1:13:02
SPEAKER_00: It's a little hard to commercialize that,
1:13:04
but it is possible to create these tools and make that
1:13:09
available to many different
1:13:13
SPEAKER_00: BCI companies
1:13:16
SPEAKER_00: and therefore lift all of the boats as opposed to,
1:13:19
SPEAKER_00: you know,
1:13:20
SPEAKER_00: throwing my hat into the ring for one more competing BCI company.
1:13:23
So I think,
1:13:23
I think this is a really like an important thing for humanity,
1:13:25
you know?
1:13:26
SPEAKER_01: So you're, you're working on like a foundational technology.
1:13:29
SPEAKER_01: I mean, yeah.
1:13:30
SPEAKER_01: And that's exciting.
1:13:31
SPEAKER_01: Um, what do you like on the.
1:13:36
SPEAKER_01: The more I cover BCI, the more I get interested and fascinated.
1:13:39
SPEAKER_01: And then sometimes I feel,
1:13:40
SPEAKER_01: um,
1:13:42
I see its limitations because I hang out with a lot of the
1:13:45
patients that get these.
1:13:47
SPEAKER_01: And so,
1:13:48
SPEAKER_01: um,
1:13:48
SPEAKER_01: you know,
1:13:49
I've covered tech long enough not to judge things in just a
1:13:54
small period of time.
1:13:55
SPEAKER_01: I mean,
1:13:55
SPEAKER_01: all these developments tend to best ones take decades,
1:13:58
SPEAKER_01: but,
1:13:58
SPEAKER_01: um,
1:13:59
SPEAKER_01: but I get a little frustrated with the limitations.
1:14:03
SPEAKER_01: I think most people have no idea.
1:14:07
that these implants do go into this tiny,
1:14:09
tiny,
1:14:09
tiny,
1:14:10
tiny part of one part of the brain.
1:14:12
SPEAKER_01: And it's very limited functionality that,
1:14:15
SPEAKER_01: you know,
1:14:16
SPEAKER_01: we're like kind of good at reading out data,
1:14:19
SPEAKER_01: but as far as I'm aware,
1:14:20
SPEAKER_01: barely not so good at reading in,
1:14:23
SPEAKER_01: although,
1:14:24
SPEAKER_01: you know,
1:14:24
SPEAKER_01: we can,
1:14:25
Ed Boyden has developed some techniques to stimulate the brain
1:14:28
and things like
1:14:29
SPEAKER_01: that,
1:14:30
SPEAKER_01: but still quite limited.
1:14:31
SPEAKER_01: Yeah.
1:14:31
So,
1:14:31
and then when I hear Elon or someone talking about,
1:14:34
you know,
1:14:35
elective consumer
1:14:37
SPEAKER_01: brain implants and going all matrix and downloading Kung Fu.
1:14:41
SPEAKER_01: I don't know, where are you on this?
1:14:43
SPEAKER_01: I mean, do you, is that like obvious to you that that's gonna happen?
1:14:47
SPEAKER_01: Do you think that's many years away?
1:14:49
SPEAKER_01: Well, I mean, you know, like I mentioned,
1:14:52
SPEAKER_00: I grew up with chronic pain.
1:14:54
SPEAKER_00: I have chronic pain.
1:14:55
SPEAKER_00: I don't know if a BCI is going to help me particularly,
1:14:59
but it's like one in three people are going to have a brain
1:15:03
disorder in some time
1:15:04
SPEAKER_00: in their lives.
1:15:05
SPEAKER_00: And so that medical need is huge.
1:15:10
SPEAKER_00: It's really right there, right in front of us.
1:15:12
And so like,
1:15:13
maybe we will get to that point as these technologies get good.
1:15:17
SPEAKER_00: But like the medical need is right there.
1:15:20
SPEAKER_01: Okay.
1:15:21
SPEAKER_01: So you're not going to sci-fi with me.
1:15:23
SPEAKER_01: You have pain from arthritis, if I can ask?
1:15:26
SPEAKER_00: I do have arthritis, you know,
1:15:28
SPEAKER_00: chronic pain is a little bit hard to track down.
1:15:29
SPEAKER_00: There are medicines that help with arthritis,
1:15:31
SPEAKER_00: but don't necessarily help,
1:15:32
SPEAKER_00: you know,
1:15:33
SPEAKER_00: when,
1:15:33
SPEAKER_00: when there are changes in brain circuitry,
1:15:36
SPEAKER_00: right?
1:15:36
SPEAKER_00: You know,
1:15:36
SPEAKER_00: there's this expression pain is in the brain and like,
1:15:39
SPEAKER_00: that's not,
1:15:40
SPEAKER_00: that doesn't mean it's in your head.
1:15:41
SPEAKER_00: It means it's like literally changes in wiring.
1:15:43
SPEAKER_00: Yeah.
1:15:44
SPEAKER_01: I get cluster headaches.
1:15:45
SPEAKER_01: Have you ever heard of those?
1:15:46
SPEAKER_01: Yeah.
1:15:46
SPEAKER_01: Yeah, it's like... Not fun.
1:15:48
SPEAKER_01: Dude, I'm in the middle of a spell right now.
1:15:51
It is like so debilitating and like the pain is just unlike
1:15:57
anything I think most
1:15:58
SPEAKER_01: people feel in life.
1:16:00
SPEAKER_01: It's, yeah, I would give anything to turn them off.
1:16:04
SPEAKER_01: And there's like pretty much...
1:16:07
SPEAKER_01: Zero medicine.
1:16:08
You have to immediately get access to some high-grade oxygen if
1:16:13
you want to have
1:16:14
SPEAKER_01: any hope of this.
1:16:15
SPEAKER_01: So it's quite difficult.
1:16:17
SPEAKER_01: So I'm with you on that.
1:16:18
SPEAKER_01: But come on, you watch Star Trek sci-fi with me.
1:16:23
SPEAKER_01: When you're just letting your mind go,
1:16:27
SPEAKER_01: you don't have any inklings,
1:16:30
SPEAKER_01: any thoughts on where the implant stuff is heading in like 10,
1:16:35
SPEAKER_01: 20 years.
1:16:37
SPEAKER_00: I think it's a little hard to predict.
1:16:41
SPEAKER_00: I wouldn't bet against it, right?
1:16:43
Whenever there are new technologies,
1:16:46
they will be used for medical purposes.
1:16:48
SPEAKER_00: But people don't just use GLP-1 agonists for medical purposes, right?
1:16:54
SPEAKER_00: They use it because they want to.
1:16:55
SPEAKER_00: And so I think there are many people who would want to take advantage of
1:17:00
SPEAKER_00: VCIs that are very powerful.
1:17:02
SPEAKER_00: And we will see what happens and we'll see,
1:17:04
SPEAKER_00: you know,
1:17:04
SPEAKER_00: what the regulatory regime looks like and so on.
1:17:07
SPEAKER_00: But, you know, many of these technologies are dual use in that sense.
1:17:11
SPEAKER_01: You said you're an optimist.
1:17:12
SPEAKER_01: I mean, are you, are you trans, are you kind of like pro transhumanist?
1:17:18
I think that there are a couple notions from transhumanism and
1:17:22
that movement which
1:17:24
SPEAKER_00: are interesting and make for really good science fiction.
1:17:27
I'm an optimist,
1:17:28
but I also very practically build technologies and I'm an
1:17:31
engineer.
1:17:32
And so I think there's interesting sources of inspiration from
1:17:38
those stories,
1:17:39
SPEAKER_00: but we need to build technologies that help people right now.
1:17:42
SPEAKER_00: Yeah.
1:17:43
SPEAKER_01: I mean,
1:17:43
SPEAKER_01: I have to,
1:17:44
SPEAKER_01: I spent so much time in AI,
1:17:46
SPEAKER_01: BCI land,
1:17:47
SPEAKER_01: all this other stuff,
1:17:48
SPEAKER_01: you know,
1:17:49
SPEAKER_01: half the days I'm so excited.
1:17:50
And then half the days I'm like,
1:17:51
oh my God,
1:17:52
the world is getting so,
1:17:54
so insane so quickly.
1:17:56
And you know,
1:17:57
sometimes it's like you're living in a sci-fi future already.
1:18:01
SPEAKER_00: Right.
1:18:01
SPEAKER_00: So.
1:18:01
SPEAKER_01: Yeah.
1:18:03
SPEAKER_01: Okay.
1:18:03
SPEAKER_01: Um, okay.
1:18:04
SPEAKER_01: I'm just keep, I'm being cognizant of time.
1:18:05
SPEAKER_01: Um, so the
1:18:11
SPEAKER_01: Taking everything you said,
1:18:12
SPEAKER_01: I mean,
1:18:12
SPEAKER_01: the thought that we will get a human connectome,
1:18:20
SPEAKER_01: it's just too TBD.
1:18:22
SPEAKER_01: It's like you said,
1:18:24
SPEAKER_01: this would be...
1:18:25
SPEAKER_01: Realistically,
1:18:26
it's dependent on a breakthrough of some sort along the lines
1:18:30
of a sequencing
1:18:31
SPEAKER_01: breakthrough.
1:18:32
SPEAKER_00: Yeah, I think the breakthroughs are...
1:18:35
SPEAKER_00: are going to be very important.
1:18:37
There was this next generation sequencing breakthrough that got
1:18:40
us to the point
1:18:41
where genomes were cheap enough that anyone could get their
1:18:44
genome sequenced.
1:18:46
SPEAKER_00: The technology is still going.
1:18:47
There's now a third generation of sequencers that are based on
1:18:49
nanopores that are
1:18:50
SPEAKER_00: potentially going to be even cheaper.
1:18:53
And so I don't know if the technology that gets us the mouse
1:18:56
brain is going to be
1:18:57
SPEAKER_00: the same technology that then takes us all the way to the human brain,
1:19:01
like whether you ride that cost curve all the way there or
1:19:03
whether there needs to
1:19:04
SPEAKER_00: be another breakthrough in the meantime.
1:19:08
But what I do think is we're going to learn a lot about that as
1:19:12
we start applying
1:19:13
these new tools that are based on barcoding and expansion
1:19:15
microscopy and new AI
1:19:17
SPEAKER_00: methods.
1:19:18
SPEAKER_00: to the problem will be able to actually very quantitatively say,
1:19:22
SPEAKER_00: how much better are we getting?
1:19:24
And then you'll be able to plot some data points on a graph and
1:19:26
extrapolate and
1:19:27
SPEAKER_00: draw some conclusions around that.
1:19:30
SPEAKER_00: And so I just want to say,
1:19:32
SPEAKER_00: The PRISM study was led by Rosa Park and Arliss Sheridan.
1:19:35
SPEAKER_00: They're some amazing scientists.
1:19:37
SPEAKER_00: Definitely,
1:19:38
SPEAKER_00: if you take a look at that,
1:19:39
you'll be able to draw some of your own conclusions and also
1:19:43
maybe email them if
1:19:44
SPEAKER_00: you have questions.
1:19:44
SPEAKER_01: Okay.
1:19:45
SPEAKER_01: Okay.
1:19:45
And then by the time this episode comes out,
1:19:48
the paper will be,
1:19:49
PRISM,
1:19:50
I will put a link in.
1:19:51
SPEAKER_01: You can go to E11Bio.
1:19:54
SPEAKER_01: You mentioned one other spot.
1:19:55
SPEAKER_01: where it will be on bio archive on bio archive.
1:19:57
SPEAKER_01: Right.
1:19:58
SPEAKER_01: Of course.
1:19:58
SPEAKER_01: And then, um, I've been to your lab in Alameda.
1:20:01
SPEAKER_01: It's very cool.
1:20:01
SPEAKER_01: You have a cool collection of people kind of like,
1:20:04
SPEAKER_01: um,
1:20:05
SPEAKER_01: it's an interesting collection of people,
1:20:07
SPEAKER_01: you know,
1:20:07
SPEAKER_01: like you mentioned a little bit earlier,
1:20:09
SPEAKER_01: um,
1:20:11
SPEAKER_01: different talents in quite specific areas.
1:20:13
SPEAKER_00: It's a really interdisciplinary team.
1:20:15
And it's a fun project in the sense that we are bringing
1:20:18
together expertise from
1:20:20
SPEAKER_00: genomics and from imaging and material science and from AI and vertically
1:20:25
integrating them and getting them to work really tightly
1:20:28
together on this problem.
1:20:29
SPEAKER_00: So I think that that is really a wonderful group of people.
1:20:35
SPEAKER_01: And how many years are you into this?
1:20:37
SPEAKER_00: So we've been doing this for about three years.
1:20:38
SPEAKER_00: Three.
1:20:39
SPEAKER_01: So are you on the clock?
1:20:40
SPEAKER_01: Do you have two left, or is there flexibility?
1:20:43
There's a little bit of flexibility,
1:20:44
but I think the sense of urgency is really strong.
1:20:48
SPEAKER_00: We've brought our prism platform from zero to one.
1:20:51
SPEAKER_00: So when we started,
1:20:52
SPEAKER_00: these ideas were out there,
1:20:55
but nobody had integrated them into a single package where you
1:20:59
can use barcodes and
1:21:00
expansion microscopy to trace neurons much more accurately than
1:21:05
you can using
1:21:06
SPEAKER_00: single color fluorescent proteins alone.
1:21:08
SPEAKER_00: So like, that's, that's a big step.
1:21:11
And then the question now is like,
1:21:13
how well does that scale as we go from,
1:21:15
you know,
1:21:15
two to an?
1:21:16
SPEAKER_01: So in an ideal world over the next two years,
1:21:19
SPEAKER_01: you would come out of that and,
1:21:21
and you would show that it can scale and you would have the
1:21:24
techniques to do that.
1:21:25
And you guys would,
1:21:26
it's fairly,
1:21:27
it'd be open source to some degree,
1:21:30
right?
1:21:30
SPEAKER_00: Yeah.
1:21:30
SPEAKER_00: So,
1:21:31
SPEAKER_00: yeah,
1:21:31
SPEAKER_00: I mean,
1:21:31
SPEAKER_00: we want to make the methods available for people to build on,
1:21:36
SPEAKER_00: especially for non-commercial research use.
1:21:39
SPEAKER_00: And so, you know, plasmids will be on AdGene.
1:21:44
SPEAKER_00: The methods will be on BioRxiv.
1:21:46
We have software libraries that,
1:21:48
you know,
1:21:48
people can use,
1:21:49
and those are open source.
1:21:50
SPEAKER_00: Okay.
1:21:51
SPEAKER_01: And since FROs do tend to have this fixed timescale,
1:21:55
SPEAKER_01: I don't think I'm out of line asking this question.
1:21:57
SPEAKER_01: I mean, do you have any idea what you would like to do next?
1:22:00
Well,
1:22:00
I mean,
1:22:00
I think everyone who is part of E11 is very mission driven.
1:22:05
SPEAKER_00: And that's why you start a new org, right?
1:22:08
Because you want to get like-minded people together who share a
1:22:11
sense of purpose
1:22:12
SPEAKER_00: and a sense of mission.
1:22:13
And so,
1:22:14
we're building this technology because we want to apply this
1:22:17
technology,
1:22:18
right?
1:22:19
If we have found an angle of attack that can reduce the cost of
1:22:23
connectomics by
1:22:24
SPEAKER_00: 100x,
1:22:25
SPEAKER_00: then we want to be able to use that to get connectomes.
1:22:29
And to continue to push that number further,
1:22:32
you know,
1:22:32
we don't want to stop there.
1:22:33
SPEAKER_00: So, you know, we're connectomics people.
1:22:37
And if you get to that point,
1:22:40
the commercial end of that would be finding cures to illnesses.
1:22:46
SPEAKER_00: It might be.
1:22:47
SPEAKER_00: I mean,
1:22:47
SPEAKER_00: I think there's a question that still needs to be answered about,
1:22:51
SPEAKER_00: you know,
1:22:52
SPEAKER_00: is what are the commercial applications of connectomics for therapeutics,
1:22:57
SPEAKER_00: for BCI,
1:22:58
SPEAKER_00: for neuro AI?
1:23:00
SPEAKER_00: All of this is sort of an open research question.
1:23:02
SPEAKER_00: And, you know, we want to, you know, help find out.
1:23:05
SPEAKER_01: Okay.
1:23:06
SPEAKER_01: All right.
1:23:06
SPEAKER_01: Well, Andrew Payne, thank you so much for coming in.
1:23:10
SPEAKER_01: It's fascinating work, like among the most fascinating, I think, around.
1:23:16
And it's been,
1:23:16
it's really cool to have gotten to know you and to hang out a
1:23:20
little.
1:23:20
SPEAKER_01: We were at a dinner.
1:23:21
SPEAKER_01: We got to shoot the shit.
1:23:23
SPEAKER_01: It was good.
1:23:24
SPEAKER_01: Yeah.
1:23:25
SPEAKER_01: Thank you so much.
1:23:25
SPEAKER_01: Thanks for having me on the show, Ashley.
1:23:26
SPEAKER_01: This was fun.
1:23:31
SPEAKER_01: The Core Memory Podcast is hosted by me, Ashley Vance.
1:23:35
SPEAKER_01: It is produced by David Nicholson and me.
1:23:39
SPEAKER_01: Our theme song is by James Mercer and John Sortland.
1:23:44
SPEAKER_01: And the show is edited by the John Sortland.
1:23:50
SPEAKER_01: Thanks as always to Brex.
1:23:53
SPEAKER_01: Anyone Ventures for making this possible,
1:23:56
SPEAKER_01: please visit our Substack,
1:23:58
SPEAKER_01: YouTube,
1:24:00
SPEAKER_01: and podcast channels to get more of what Core Memory makes.
1:24:03
SPEAKER_01: Thanks, y'all.