Hey, everyone. Good morning. I'm Tejas Sawant. I cover the life sciences at Morgan Stanley. Before we begin, important disclosures, please see the Morgan Stanley Research Disclosure website at morganstanley.com/researchdisclosures. If you have any questions, do reach out to your sales rep. It's my pleasure this morning to host Ginkgo Bioworks, and speaking on behalf of the company, we have Jason Kelly, CEO.
Yeah, happy to be here.
Thank you so much, Jason, for joining us today. Maybe, just—
I'm filling in for Mark.
Yeah, I know.
Opportunistic travel change, so I'm glad to be here.
Yeah. Maybe, maybe just to kick things off, Jason, could you share with us, you know, the key, key accomplishments that you're really proud of in terms of Ginkgo's journey this year? It's been a tough year, lots of macro headwinds and cross currents in the portfolio, which we'll get to.
Yep.
Talk to us about, you know, what you're most proud of in terms of the company and the organization, and then we'll go from there.
Yeah. Yeah, so I'll give a—I can give just a little bit of general context, too. We went through about a 35% RIF earlier this year as part of a general cost takeout for the company. Not an easy time, you know, we're obviously very thankful for the employees that got us to where we are today. A big part of this was really to focus on getting the costs in the company, particularly for our solutions business.
Mm-hmm.
Teja, I know you know this well, but, you know, the history of Ginkgo is we have been engineering cells on an R&D partnership basis for companies from, you know, Bayer CropScience in the ag space, Novo Nordisk, Pfizer, Merck, on the biopharma space, and also industrial biotechnology companies, like Givaudan.
Mm-hmm
-in the fragrance industry.
Mm-hmm.
And in all those cases, we call those solutions deals. They're basically R&D partnerships. We get some fees, and we get some downstream value share. Could be milestones, it could be royalties.
Mm-hmm.
All right, and what we don't do is we don't have our own products.
Right.
We don't have our own pipeline.
Right.
A lot of the history of Ginkgo has been figuring out, is there a more of a tech platform style business model for advanced technologies in biotech R&D?
Mm-hmm, mm-hmm
That isn't make a pipeline of drugs? All right, and if you look, you know, like a company like peers, companies like a Recursion or something, would have a-
Mm-hmm
Highly automated, AI forward, like technology platform, but the way that they're commercializing it is they're going to develop their own drug pipeline.
Mm-hmm.
That has been a proven way to commercialize these things. We believe at Ginkgo that you could get them to market with, again, a services style business model, and that in the long run, that creates the kind of feedback loop that drives something like what we've seen in the tech industry.
Mm-hmm.
Where you get compounding technological improvements, continued investment, and ultimately really huge market cap, like industry-changing companies.
Mm-hmm.
All right, so that's why we did it.
Mm-hmm.
What's happened in the last year? I think with our solutions business, we sort of started to hit a limit of a mix of customers being willing to outsource that sort of thing, and also importantly, outside of our first markets, which was industrial biotech, the timelines to getting to royalties and milestones
Mm-hmm
-for biopharma in particular, are so much longer, that supporting the whole company on the back of milestones and royalties that were in the future, was not really compatible with capital markets and just generally, like, where we were spending. Mm-hmm.
What we are doing in solutions, and I am quite proud of this, is cutting costs down so that we can support that work on the fees we receive from customers.
Mm-hmm
As close as we can.
Mm-hmm. Mm-hmm.
Those milestones and royalties, they will come—
Mm-hmm
When they come, we'll still be here to get them. All right, so that's really what we're doing on solutions. That was where a lot of our effort was in cost takeout in the first part of this year.
Got it.
What I'm happy to talk about next, we can get into this, is where are we investing for growth?
Yeah.
That's really on the tool side, and it's more traditional services businesses, like you would think of a traditional CRO.
Right
but doing different things we can talk about.
Right
And equipment, selling some of our robotics.
Got it.
Yeah, happy to get into that. I would say that cost takeout and that focus on the solution side, no small feat, and you saw last quarter, you know, I was quite happy with where we were on revenue.
Mm-hmm.
Even in the midst of all that change, the team really pulled through and continued to deliver for our long-term customers. Customers we have a lot of repeat business with, like Novo, like that, that's been really great to see.
Got it. Let's begin a little bit, start with cell engineering. I think you've got about 140 current active programs there.
Yeah.
Food and ag, largest, followed by pharma and biotech. Just as a point of clarification, is the revenue composition similar to the program count, or, or—
Yeah
does it vary, sort of significantly?
Yeah, it's not, it's not too far off.
Okay.
And again, I won't belabor this, and maybe I'll y ou know, I know, I'll probably take us off script a little, if you don't mind. The, the I would say on the solutions business, we're probably about, you know, a quarter to a third-
Mm-hmm
-pharma, a quarter gov, and then the remainder is, is probably a- it's ag and industrial, but like majority ag.
Okay.
Okay? So that's kind of like how it breaks down. I think one of the things that's cool about the solutions business is we have applied it much wider than anyone else. Like, you might look at like a company like a, Adimab-
Mm-hmm
-for example, has a very similar business model. Fees, milestones on the back end.
Mm-hmm.
What do they apply it to? They apply it to antibody binding discovery, basically.
Mm-hmm. Mm-hmm.
You know, they don't go off and help with, plant trait discovery.
Right
O r biologics, microbes in agriculture, or animal feed.
Right
Like, all, all these other markets.
Right. Right.
Like, so I think Ginkgo has done a nice job showing that, and these are not easy deals to get. We can convince customers that we have something special in the area of research, to develop a new microbe or mammalian cell or a fungal cell against their application.
Mm-hmm.
It is great. I think the spread is important. What I think is, I think I'd love to chat with you about if you're up for it.
Mm-hmm, mm-hmm
I s sort of where we're headed on the tool side.
Yeah.
Okay.
Yeah.
Does that work?
We'll get there.
All right, fine. Okay, we'll just go down the list.
Yeah. You know, I want to talk to you about the narrowing of the focus on the cell engineering side.
Yeah.
Now it's gonna be pharma and industrial solutions, and ag. Can you just clarify, like, what percentage of your business is this new subset of offerings? And then, you know, as you think about how this new structure positions you better versus the all-comers approach you'd taken earlier, talk to us a little bit about that, because I'm sure it involves a little bit of, you know, saying no to-
Yeah
Sort of paid work, right?
Yeah. This is a good point. One question is like, how do you serve the solutions?
Mm-hmm
B usiness-
Mm-hmm
with less spending?
Yeah.
Right?
Yeah.
Like, wave a magic wand, right? The answer is, the way we previously would sell solutions, we'd walk into a customer, a Merck or something, and it's sort of like: What problem do you have?
Mm-hmm.
Like, you know, give me, like, a high technology problem you have in the cell engineer—you know, like, it, it wasn't like chemistry.
Yeah.
We're not a chemistry company.
Yeah.
If it was in biotech—
Yeah
and cell engineering, what do you got?
Mm-hmm.
Right, and if it's Pfizer, you know, I need this RNA to be more stable. Like, whatever it is, right? And we would say, "Great, let's, let's take that back." We'd sit down with our scientists at Ginkgo, look at all of our infrastructure, put together a work plan, bring it back to their scientists at the customer. "Oh, I like that work plan.
Mm-hmm, mm-hmm.
Okay, great.
Mm-hmm.
It might be a thing we had not done that exact kind of work before.
Yeah.
It was gonna use pieces of our infrastructure.
Yeah
B ut we hadn't, like, really run that whole workflow previously. And so that meant that there was gonna be a whole set of effort at Ginkgo.
Mm-hmm
T o sort of, like, get that really, like, mainly to a high throughput. That's usually what we spend the most time doing, is, like, getting it really bulked up in high throughput.
Mm-hmm, mm-hmm
B efore we could start the program, before we could start the program in real scale. Getting the customer to pay for all that, in addition to the running of the program.
Right
T hat's a tough sell on the fee side.
Right.
You could do it and say, "Hey, this is a great area. We think there's ten more deals behind it.
Right.
We're happy to have the customer pay for any of it on the fee side. We'll make some money on the, on the royalty, the milestones.
Right
T o make the unit economics of the deal good. Plus, it's essentially covering some of what is really tech development for me.
Yeah.
Does that make sense?
Yeah.
That's the kind of deal I wouldn't do anymore.
Got it.
All right?
Yeah.
Because that is essentially me putting y ou know, we have, we, we, I mean, look, one of the things I think we're happy we did is work on the cost takeout while we still have a large-
Yeah
cash position.
Yeah.
You know, we ended the second quarter.
Yeah
W ith $730 million.
Yeah.
We have no debt. Like, you know, like, we have money for growth capital, it's a question of what do you want to deploy it on?
Right.
I don't want to deploy it anymore on doing a new solutions thing that I haven't done before, because the customer is asking for it.
Fair.
That does reduce the scope of what I can sell.
Got it.
Because we've done such a wide variety of things previously, I actually have quite a few things I can sell.
Got it.
We're still, we have a more narrow set, but yeah, absolutely, it's in industrial, it's in biopharma and ag, but it isn't anything in those areas.
Yeah, fair enough.
It's more consolidated around previous work we've done.
One follow-up there, Jason. In a sense, the key differentiator for Ginkgo has been that Code base, right? And it's not always about what works, it's also about reducing the dimensionality of the space for your customers or your marginal customer by telling them, "Okay, here's the eight things that we definitely know are not going to work.
Yeah.
Right? And that was largely a creation of that all-comers approach.
Sure
A nd then sort of going down those rabbit holes, right?
Uh-huh.
So does that now sort of get, a little bit deprioritized, if you will,
Sure
as you try to navigate the macro?
Yeah, I'll give you a couple comments on that. The two core assets of Ginkgo have always been our Foundry and our Codebase.
Yep.
Part of what the customer is coming for is also access to essentially high-throughput biotechnology.
Right
I n, for like, sophisticated lab work.
Right.
Okay? Right, like, "I wanna run this one screen a million times.
Mm-hmm.
There's a lot of ways to do that.
Yeah.
Once you start to say like, "Oh, I wanna do these nine antibody developability assays at the scale of thousands," it's complicated, okay? That foundry, I don't want to undersell that. That was part of the reason.
Yeah
C ustomers were coming. The other reason they're coming is, like you said, Codebase.
Yeah.
Codebase is sort of our catch-all term for the know-how, the intellectual property, the, like, learnings we accumulated as we did-
Mm-hmm, mm-hmm
these cell engineering projects.
Mm-hmm.
All right? Importantly, when we would do those solutions deals, we would also retain the reuse rights.
Yeah
F or all that Codebase.
Yeah.
Okay. That was extremely frictional.
Right
L et's be very clear. And I actually think, like, we needed a little bit of that, because some of the absolutely, like, most core Codebase we were developing in the course of customer projects.
Mm-hmm
A nd we knew that was gonna be reused a lot.
Mm-hmm, mm-hmm.
We wanted to ultimately make it available to other people.
Mm-hmm
F or exactly the reason you said.
Mm-hmm.
Like, we know where the bodies are buried.
Yeah.
We've tried it before.
Yeah.
We know that doesn't work.
Right.
Let's not repeat it for you, let's just incorporate that-
Right
into the next project.
Right.
All right. You, at the same time, as the customer, are like: "Whoa, you know, don't give anything to any of my competitors who are gonna help him out if I paid for it.
Right.
Which I actually think is reasonable.
Right.
I think we kind of made a mistake there. We pushed it too long, hanging on. The change, I think, is basically permanent. I mean, I never say permanent, but like, you know, close enough, is we, you will not see us demand broad reuse on IP-
Fair enough
for customer projects. You will see us, if we think a piece of Codebase is very general, we'll just develop it ourselves.
Got it.
All right? So, so that, that's the, that's the short answer.
Yeah
O n the, on the Codebase side.
Makes sense.
You will see us now on the tool side, though, find interesting ways to start offering Codebase to customers-
Mm-hmm
t hat previously we would've locked up. Okay, so in general, on the tool side, what I'm gonna do is kind of democratize and open up this platform-
Mm-hmm.
you know, both the Foundry and the Codebase
Mm-hmm.
that was previously only available to Ginkgo scientists doing a solutions deal with you. I want your scientists, again, at, you know, Bayer or wherever
Yeah.
-right? To just be able to access that themselves.
Right.
I've got a bunch of ways to do that, including for the Codebase.
Fair enough. So, you know, as a follow-up to that, Jason, I mean, the customer perception around data sharing—
Yeah
Takes time to change, right? I mean, we've seen a few examples in the life science—
I don't think it'll ever change in life science. It is-
Oh, I meant more from the perception of where, you know, if a vendor has been keeping the data—
Okay.
-they pivot to not keeping the data anymore.
Yeah
P eople still think of that vendor as someone who was hoarding all the data.
Oh, I see what you're saying.
Right?
Okay, sure.
Where are you on that journey? Because it's one thing to put out a press release saying, "We're not gonna do it anymore.
Yeah.
It's another thing my customers sort of wake up.
Sure
-to that reality.
Yeah.
And then second, you know, on the tools program, just walk us through the strategy behind that decision and why now?
Yeah. Yeah, so to your first question about like perception and then, and just remind me the second one in a second. So the I do think the perception thing is a key point, actually.
Mm-hmm.
Because we had this experience, we recently announced, and we're kind of like soft launching this, that you can buy the robotics directly from Ginkgo.
Mm-hmm.
Right? This is some of the technology we acquired from Zymergen, and then over the last two and a half years, we've really tuned it up a lot. We have, like, new versions of all the hardware and everything. Like, we put a post up, like on LinkedIn or something, and a bunch of comments are just sort of like, "Wait, is this in your facility or my facility?" You know, right? Like, because the impulse is like Ginkgo's stuff has been
Right
-kept tight.
Right.
Right?
Right.
It is good and bad in the sense that, like, when i t does, it is like a thing that we have to explain to people.
Yeah
because it's not their expectation.
Mm-hmm.
People are excited about it too, because it's a little bit like getting to go in Willy Wonka's chocolate factory, you know, right? It's like, "I've seen it, but I don't know what's in there," you know, right?
Yeah.
Like, the s o in that sense, that actually is creating demand for us, too.
I see.
Right?
Yeah.
Because people are sort of saying like, "Well, great." I mean, maybe, you know, "I don't know why you didn't do it sooner," and we probably should have.
Right.
I think they're generally happy to have it, so.
Okay.
I wanna be clear, like, you know, Ginkgo is a very mission-driven company.
Mm-hmm.
The mission of the company is to make biology easier to engineer.
Mm-hmm.
That's, for example, why you don't see me having a drug pipeline.
Mm-hmm.
Right? Is because my mission is not to cure cancer.
Mm-hmm.
Right? Other companies. I think it's an important mission.
Mm-hmm
This is not my mission.
Mm-hmm.
Right? My mission is to make biology easier to engineer, and I'm trying to find the way that I can do that with the biggest impact.
Got it.
Early on, we pretty much had to do solutions, Teja, because we were the ones who really believed that there was a much better way to do the biotech research if you adopted automation, if you adopted high, you know, large quantity data science.
Mm-hmm
-like all these things, we believed it-
Mm-hmm
ML, more than other people.
Mm-hmm. Mm-hmm
If I signed an R&D deal up with you, the good thing was I got to decide how to do the work.
Right.
That meant we could do it our way.
Right.
Okay, we're, you know, we did Y Combinator ten years ago. We're like ten years into a scaled version of that, and I have a lot of proof points now.
Right
that this is actually a better way to do it, and the world has changed.
Mm-hmm.
Like, look at all these like AI bio companies that are basically like, "Hey, we're data limited. We need to generate these large data." I mean, that is a story that like, that dog didn't hunt.
Right
Ten years ago in biotech, right?
Right.
Like, people just would not spend that kind of money. They didn't really wanna generate the data, didn't believe it. And it's still early in that—
Mm-hmm
It's come along enough that when I put these tools out there, there are people that want them.
Hmm.
The only people that really wanted them ten years ago, that really believed that there was a totally different way to do this stuff, was like, us, right?
Right.
Like, we had to be our own customers of the infrastructure. Does that make sense?
Yeah.
And I'm, I'm like overstating that, like we had- there are fellow weirdos out there- but like, but, but that was what it felt like, and that, that was a big part of the reason we started with solutions. But it's a total mission set.
Got it.
I'd love to get everybody using our robotics.
Got it.
I'd love to get everybody access to our Codebase. That means there's more bioengineers.
Right.
That means biology is being made easier to engineer, you know, right? Like, I don't have a problem with that. It's great.
Fair enough. Does the program—
Also, people will make a killing.
Sort of expand the addressable market? And what are your, what are your expectations for, you know, revenue growth from tools in, in, in sort of the years ahead?
Yeah. Okay, so yeah, so finally, you finally let me get there. Okay.
Cool.
All right, so let me tell you about the tools. All right. So, short form of it is we want to take the platform we've already built.
Yeah.
Again, on the solution side, I'm trying to, like, get costs under control there, make sure we can deliver that, like, at a rate that is, like, paid for by the fees, and then make money in the future on the milestones.
Mm-hmm
-and royalties. Keep signing those deals.
Mm-hmm.
Fine. Like, where I wanna invest growth capital is I wanna say, "Hey, here's our platform.
Yeah.
I've already got it built. Let me build the channel to sell it to your scientists directly.
Got it.
I don't need to reinvent a whole crazy new thing. I need to take what was already being used by Ginkgo scientists.
Mm-hmm
I need to actually make it something you can buy.
Got it.
All right, does that make sense?
Yeah.
That does take some work.
Yeah.
Right?
Yeah.
That's not for free.
Yeah.
Right. Okay. And, and so I'll give you a couple examples. On the robotics side, we have technology. The big problem in, in integrated robotics—so, so let me just explain robotics. No one uses robotics in life science.
Mm-hmm.
That's the first point, okay? It is, it is thinly used. You have lowest level of that automation is like walk-up automation. A company like Hamilton has a liquid handling robot.
Right.
You stand in front of it as a scientist, you put plates on it.
Right
Y ou interact with their custom software.
Right
you tell it to do something.
Right, right.
You come back later and take the plates away.
Right, right.
What do you take the plates away to? Maybe you take them over to this plate reader.
Right.
-that you buy from Thermo Fisher, or they go to this mass spec or whatever, right?
Right.
The mass spec has a little bit of automation built on it.
Right.
Like an auto sampler or something.
Right.
Okay, right?
Right.
It's sitting there. Okay. Like automation, that's like, like-
Sporadic.
Yeah.
Yeah.
Okay.
Yeah.
The integration of all those things, of all that equipment in the lab, of course, is you.
Right.
You, the scientist.
Right.
You walk around with the plates, and you bring them to the different equipment, and
Yeah
-you are the integration. All right, so integrated lab automation, you can get from a company like HighRes Biosolutions or the high-end of Thermo Fisher Scientific's automation platform. There's like an arm-
Mm-hmm
in between a bunch of equipment
Mm-hmm
-moving the plates around.
Mm-hmm.
It's pretty cool. Okay, right? So now suddenly, you can run this machine overnight.
Right.
You can move many more plates, right?
Right.
It doesn't make mistakes.
Right.
Like, okay, oh, that seems good.
Mm-hmm.
Trouble is, those setups are built custom.
Mm
-for whatever specific thing you wanna do.
Mm-hmm.
I have these six pieces of equipment, and I wanna move plates between them in this order, and I wanna do that day and night, and okay—
Mm
Here we go, and I'm gonna make you a whole system to do it.
Right. Mm-hmm.
Here's the problem. Six months later, one of your scientists is like, "I just read this paper. There's a much better assay for what we were doing.
Mm
I need this new piece of equipment, and we need to change the protocol.
Right.
Oh, my God! I just spent $8 million on that integrated automation setup. It took a year to build it, six months to teach everybody to use it.
Right.
You wanna change it? That is a six-month project.
Right
To add a new piece of equipment, reprogram it, retrain everybody. Okay, right.
Yeah.
As a result, knowing that-
Mm-hmm
People barely buy it.
Mm.
All right.
Yeah.
It's not future-proof.
Mm-hmm
-to changes.
Mm-hmm.
Okay, and then you only end up really using it in much more constrained automation environments, where you know you're gonna use the same thing for, like, five years.
Mm-hmm.
Does that make sense?
Yep.
All right. We suffered this exact same problem at Ginkgo, 'cause we were deploying automation, our scientists were changing their minds on what they wanted.
Mm-hmm.
We had our own in-house engineering team and everything else, and we were still having to change things all the time.
Mm-hmm.
Painful. All right? The folks at Zymergen, same problem. They invented technologies proprietary to Ginkgo. We-- It's hardware. We OEM manufacture it. We do final assembly in Emeryville, and it is a cart-
Mm-hmm
with a piece of equipment, and a robotic arm, and a magnetic rail that allows you to move samples down a little track.
Okay.
A little train track.
Yep, yep.
You have a cart, you have a cart, you put them together, and the rails connect.
Got it.
All right, and the trick is, if you have those six pieces of equipment, I sell you six racks.
Mm-hmm
-with the equipment in them.
Mm-hmm.
We put them in your lab, they have the rail and the loop.
Mm-hmm.
You can start doing that original idea your scientist had. Now it is six months later, and they're like: "Hey, yeah, I gotta change-
Mm
-one of the pieces of equipment.
Swap out.
I sell you another rack.
Yeah
-with that piece of equipment.
Yeah.
We pop it on the rail.
Mm-hmm.
A week later, you're running the new protocol.
Got it.
All right, and here's really cool. You have a, now another protocol you just invented that uses the same seven pieces of equipment, but is a different order.
Mm-hmm.
Interleave it. You can have the software add that workflow to the first one, and now you're running both of them on the same thing instead of buying a whole second unit.
Interesting.
Okay? How about you want me to prototype a protocol for you because you do not have the automation engineers? If I have the same seven pieces of equipment in Boston, I can prototype it.
Mm-hmm.
You have them at your place, I just send you the updated workflow, and now suddenly your thing is running it tomorrow, and I did all the work to qualify the biology for you.
Got it.
Okay, so this is a different paradigm for deploying integrated automation. I think it's a really cool thing. I think it's actually better in everybody's hands than just in Ginkgo's hands.
Mm-hmm
for a whole bunch of reasons. And so we're super excited about it. We just started offering it. We just announced it last week. We've been kind of soft selling it for the last three or four months. The good thing is, Zymergen had actually sold a couple-
Right
two and a half years ago. We have been already serving external customers.
Mm-hmm
-in National Lab, as a startup biotech company-
Mm
called Octant. That's. A nd so we have the experience of serving as an external vendor, you know, repairs, so all that stuff, we've been doing that for the last two years.
Got it.
I think this is one that I think we just really need to scale the sales channel.
How proprietary is this?
Very.
Interesting.
Yeah.
A customer who sees this can't just sort of reverse engineer and just start building it on their own?
Number one, it, it's like an actual custom hardware job.
Yeah.
You have to choose-
Yeah
-to get into the hardware business, which maybe you could if you're a tools company-
Right
A customer's not gonna do that.
Right.
We have a whole software stack.
I see.
Which, you know, Zymergen started working on this eight years ago.
I see.
Yeah. So, like I think it's pretty y ou know, and we have patents, you know—
Mm-hmm
-all the things. Like, I'm sure there's other ways to get at this problem.
Yeah, right.
I think we are well ahead on having a productized thing here, both in terms of the hardware and the software.
I see.
Yeah.
Explain to us the monetization of this. Is there an upfront sort of—
Yeah
CapEx outlay, and then there's gonna be a license fee of some sort?
Yeah. We're in it. This is what we're figuring out right now.
Yeah.
We're engaging with a number of customers.
Yeah
to get our arms around this. I think you can imagine a few different approaches.
Yeah, yeah.
A, it's equipment.
Yeah.
You can do a, you know, capital expense.
Mm-hmm.
People can just buy it that way. If you want to engage with smaller companies, you could like lease it or maybe make the equipment free-
Right
-just charge for the software.
Right.
Either way, you're, you're gonna pay for software and services on top.
Yeah. Mm-hmm.
Like our current customers already do that. So you'll have some recurring kind of SaaS revenue that way as well.
Mm-hmm.
We're flexible. Right now, we're engaging with sort of first customers and sorting that out.
Is the idea now, over the next sort of, you know, six months, let's say, to set up some marquee customers who are running this?
Yep
A nd, you know, have them sort of, in a sense, you know, market it for you more broadly?
Uh, yeah.
Okay.
Yeah, I would say that's the plan. I mean, in general, I mean, the good thing about Ginkgo, like, if you think of this as like a sort of, again, I'd say the technology is quite mature, but going out and selling it, it's in the early stages, a little bit like a start-up. The good thing is, I mean, Ginkgo is excellent at enterprise.
Right.
Ginkgo is excellent.
Right
-at deal team. Like, we're good at moving contracts-
Mm-hmm
We have all that other infrastructure in place. It is really a question of, like, sales channel and uptake.
I see.
Yeah.
Fair enough. I wanna talk a little bit about lab data as a service.
Yeah.
You know, you launched that offering, you know, fairly recently as well at Ferment.
Yep.
Talk to us about the value proposition and, and particularly, you know, as the use of AI/ML approaches go mainstream in drug discovery.
Yep
How is it resonating with, with your customer base?
Yeah. So the idea here, we're actually gonna call this Datap oints.
Okay.
The idea. And so great example, AI bio company wants to generate a large data set to train a model.
Yep.
I'll give you an example of the kind of data you might want. There's a lot of ways to find an antibody binder-
Mm-hmm
-out there, right?
Mm-hmm.
Like yeast display.
Yeah
-and hit a mouse or whatever, and you have companies that'll do it for you.
Yeah
-like Abcam and whoever. There's n- then you get your hits-
Mm-hmm
-and you go send them off to WuXi or somebody to see, like, to run them through a set of developability assays, or you do them in-house.
Mm-hmm, mm-hmm.
You're basically looking to see things like solubility. Like, is this gonna be a good drug?
Right.
I know it's a good binder.
Right
Is it actually gonna be something?
Right
-tolerated?
Right.
Okay.
Right.
There are, like, call it 10 assays you wanna do there. There are, depending on which one it is, they can be pretty challenging assays.
Yep.
The way you do it now is you kind of cross your fingers and hope the ones that were good binders are also good, well, good candidates on the developability axis. If they're not, you could go grab other ones out of your pile, but you don't have a really great sense of, like, which ones are gonna be the ones that end up
Right
-like, passing muster.
Right.
Does that make sense?
Right.
There is a lot of interest on the AI side, is, could we actually use an AI model to predict whether that binder is gonna be good on the developability assay?
Mm-hmm.
Where's the data for that?
Mm.
'Cause everybody's only got these, like, small quantities of developability data.
Mm.
That's one where, well, what I'd actually love to have would be 10,000 or 100,000 Datap oints.
Right
about different antibodies
Right
-on these ten developability assays. That is an example of, like, one of our first products in Data points-
I see
-is antibody developability at scale.
I see.
If you want a couple of them, go call WuXi, but if you want 10,000, you call us.
I see.
Right? Same for functional genomics.
I see.
Right. Things like, you know, Perturb-seq, Drug-seq.
Mm.
You want to basically make edits to mammalian cells.
Mm-hmm
-with CRISPR, or you want to hit them with a compound library, and then you wanna run a set of high content omics assays on them. Tell us what you want.
Interesting.
Right? We can again, generate that data for you. It goes back to your data science team, like-
Yeah
n ot in spreadsheets, right? Like—
Right.
you know, like it's, it's done.
Right
Cleaned up. Right. And so, because we have in-house, on the solution side, been work- we've been doing large data set gen and ML work and everything else-
Right
-for five years, you know, right? Like, like, you know, we have all this stuff.
Right.
And so it's really that exposing it, pricing it, it's gonna be very trans-- you know, the pricing is transparent, like a menu.
Mm-hmm.
Right? Eh, but where you would call Ginkgo is if you need a lot of data.
I see.
Okay. If you need, if you need an outsourced set of hands, you know, we're not your guys.
Got it.
Okay.
So-
That sort of Data point.
And-
We want, and you could see that going to many other types of data in the future.
Right
-too.
Right.
It's a general idea, right? It's sort of like what da- but the question that we're trying to sort out right now is where, where do we see customers asking us for, like, large data sets?
I see.
Yeah.
Fair enough.
You own it, by the way. All that data belongs to you, IP belongs to you. Pure fee for service business, right?
I see.
We'll just make money by doing it cheaper than we charge you. That's it. Simple as that.
Got it. Got it. You know, on that note, Jason, I want to run this, this Nature Communications paper that came out recently talking about, you know, AlphaFold not being able to deliver as promised. It failed to correctly predict both protein structures in about, like, two-thirds of the proteins where it had training data, and I think most of the proteins where it did not have training data.
Yep.
Disappointing to see. Can lab data as a service help improve upon that for a customer who's using a data set or an algo like AlphaFold?
Yeah. So a couple areas. Let me just speak to, like, AI-
Yeah
-and where, like, better tooling could potentially help. There's sort of two lab activities associated with an AI model.
Mm-hmm. Mm-hmm.
One is the lab work that generated the data that you use to train the model.
Mm-hmm.
Frankly, the biggest data sets are public data sets, where that data was generated over years.
Right
L ike the Protein Data Bank.
Yep
-or the-
Yep
-or GenBank-
Mm-hmm
Dispersed across many labs, data collected in an organized way. Those are kind of the big ones, to be honest, PDB and GenBank. Part of the reason is, it's easy to measure that type of data in a way where you trust that someone else who did it is giving you decent data.
Hmm.
It's because, like, everyone kind of agreed on standards a long time ago.
Mm-hmm
-around the protein stuff, and it is like just basically like a atomically physical, like a shape.
Yeah
-and a sequence to sequence.
Right.
If instead you wanted data on, say, like mRNA stability in a certain human cell type, and this guy did it in his lab, and she did it in that lab—
Mm-hmm
-and you try to put them together, you don't believe that they were done the same way.
Right.
There's-
Right
Much less data produced, distributed across labs for training.
Mm-hmm
-beyond PDB and GenBank.
Hmm.
And I'm, I'm oversimplifying—
Yeah
That's a little bit of the gist. One of the places that lab data as a service, or Data points as we're calling it, could help—
Mm
-is make me from one place, a bunch of known comparable data for training.
Hmm.
Okay, right, and it depends what you want.
Yeah.
Maybe you want antibody developability, but someone else wants mRNA stability.
Right.
-and somebody else wants this, and someone else, you know, like-
Right
Great! Someone wants that data in primary, you know, whatever, neural cells.
Right.
Right, like, you know, like-
Right
like, it just depends what they want.
Right.
Right? But the data sets aren't available publicly at the scale they need, because in order to create the scale, you have to jam together a bunch of data sets, and you don't actually—they're not really comparable.
Got it.
Does that make sense?
Yeah.
That's one area.
Mm-hmm.
Then the other thing is, okay, I want to, what's called fine-tune my model.
Mm-hmm. Mm.
You might be familiar with this, like, in OpenAI Land and in the English language models. There's now companies where you can pay a company, they'll take all of the Morgan Stanley in-house documents.
Mm-hmm
T ake GPT-4, the-
Yeah
-thing trained on the whole internet.
Yeah
-but really feed it all the Morgan Stanley documents, and it really learns everything. Then when you ask it a question about company policy, it does not pull company policy from Wikipedia.
Right.
It is mostly learned company policy at Morgan Stanley.
Right.
Does that make sense?
Right.
That's called fine-tuning.
Yeah.
In order to do that, you need a bunch of relevant data, and you need to kind of cyclically teach that model, particularly as you generate more of it.
Mm-hmm. Mm-hmm.
Same idea holds here. You could have a really generic model like an AlphaFold, and you would say, "Well, fine, AlphaFold knows about every protein structure under the sun, but I care about antibodies.
Hmm.
I'm gonna. People have even done this already, at David Baker's lab and others.
Mm-hmm.
I'm gonna take antibody data.
Mm-hmm
I'm gonna really—
Hmm
fine-tune it with just that, like giving it just the Morgan Stanley documents.
Right.
Not every document in the world. Remember, it learned originally on every document. That's why it speaks English.
Right. Right.
You're like, "I really want you to pay attention to these.
Right.
Does that make sense?
Yeah.
Same idea here. In that scenario, you could use Data points to generate protein data in your area of interest and use that to tune up AlphaFold.
I see.
Okay, right, or in this case, folding is a little trickier because the folds are—
Right
Hard to do at high throughput, but like other protein properties.
I see.
Okay? That in general, I think is also an area that would help. Last but not least, our Code base. We're going to try-- we're going to take a lot of these proprietary large datasets we already have at Ginkgo, embed them in models-
Mm-hmm
-and make those available to people, too. Okay. Again, just no IP, nothing else, fee for service. Go have fun.
Got it.
And so, so those are some of the directions you'll see us try to nudge things along. But we're, we're not like, you know, we're not an AI bio drug developer.
Right.
We think that's a great area.
Right.
I'm hopeful that it-
Right
Revolutionizes drug. I think that's great, right?
Right.
What we really wanna do is provide tools to people that need large datasets.
Got it.
Yeah.
Fair enough. All right, almost out of time, but I do wanna run a quick numbers question by you on the biosecurity side, right?
Yeah.
Recently, there was news around that Traveler-based Genomic Surveillance program. The contract was about $94 million or so. =
Yeah.
Was that the same thing that you guys were, you know, referring to earlier? You had talked of, you know, potentially getting a CDC contract.
Yeah. So, again, I can't speak about this too much publicly just because it's a government-
Mm-hmm
-project, and so everything has to be done together. Yes, there was a contract posted on a government website that has to report out on newly signed contracts-
Got it
for traveler genomics. And our program with the CDC and the TGS program in general, if you remember, is the collection of wastewater.
Mm-hmm
-from airplanes.
Yep.
And then you look for viruses, and then you sequence them if they're there.
Got it.
It is pretty neat. It is not just I mean, it obviously got started during COVID.
Yeah
B ut like, you know, mpox, H5N1.
Yeah
R ight? Like, like, and I'll just say that the whole point of this is to have a radar system, like we monitor for the weather.
Mm-hmm
-for things that are a lot more dangerous than the weather. So it really feels like this is something we should have in place. I'm happy to see that that got put on that website.
Got it. So what's the cadence of the revenue recognition for you guys? It's like a multi-year period, or is it sort of quicker than that?
Yeah. In general, with our I mean, I'll speak to biosecurity generally.
Yeah.
I think what's cool, what's great about that contract, as you see, it's like over a three-year period.
Okay.
One of the key things for us in biosecurity has been going from like, the episodic revenue of COVID.
Sure.
By the way, I'll just l ike, I think the way biosecurity revenue will end up looking is, you'll have a baseline of constant monitoring.
Mm-hmm. Mm.
Airports, other places, you know, maybe animal facilities, like, places where disease is born.
Mm-hmm.
Then episodically, things will happen, and when that happens, you surge against it, right? If there's like a, like mpox right now-
Yeah
In Europe, that could be a surge, right? And then you need to go and actually suddenly monitor much more aggressively because you're trying to tamp something out. That's my guess of how that ends up looking as well.
Got it. Fair enough. One final question, and we'll get you out of here. You know, I don't know when it was, probably like five, six years ago, Jason, I remember speaking with you about, you know, this, the approach of a horizontal SynB io business model.
Yeah.
You know, there were a couple of, like, failed examples of companies that tried and sort of, you know, not, not e ssentially ran out of runway.
Yeah.
I remember you saying something to me then, that they weren't wrong in taking that approach, they were just too early.
Yeah.
You know, as you look back at, you know, some of the challenges that you've had at Ginkgo over the last, you know, year, year and a half or so, has your conviction in that approach evolved at all?
No, I still think it's the right thing. Yeah, horizontal is the move. It is, it is an absolutely great question. Yeah, the, if t he problem with vertical is you can't do more than the product that comes off-
Mm-hmm
that one product that comes off.
Mm-hmm. Mm-hmm.
If you look at, like, the other great engineering fields, and, like, deep in my heart, I know that bioengineering runs on the same physics as everything else. It happens to have code inside cells.
Mm-hmm. Mm-hmm.
It should match what we've seen in other engineering fields. The greatest companies are the platform companies. I think even if you look at tools companies like Thermo Fisher, which is really our greatest tools company, $200 billion plus market cap company.
Mm-hmm
They're the horizontal platform for working at the bench.
Mm-hmm. Mm-hmm.
Okay? That's what they sell you. I wanna be the horizontal platform for working at high throughput in robotics, right? Like, for these high-content, high-volume data.
Got it. Fair enough. Great place to leave it at.
Yeah.
Thank you so much, Jason.
Thank you.
Appreciate it.
Yeah, appreciate it.
Yeah.
Good to see you.
Yeah, of course.