COSMOFACTORY

Deanna: [00:00:00] This episode is about molecular design. It's about creating proteins with the help of artificial intelligence. About the convergence of efficiency, sustainability, and technology. It's also about bioreactors. Digital twins and accelerated innovation. My guest on today's episode of the CosmoFactory podcast is Lucile Bonnin, head of personal care at Cambrium.
Lucile, welcome.
Lucile: Thank you so much, [00:01:00] Deanna. I'm very happy to be here today.
Deanna: Yes, yes, you're welcome. And I'm sure we'll have a very exciting and inspiring interview. Um, Just to help us locate our conversation today, um, let's say that Cambrium is a cosmetic ingredient company, uh, specialized in novel proteins, um, and let's say that this company is also, if you will, um, an AI based biotech research company.
So, putting these ideas together, um, Cambrium develops protein molecules, uh, for use in cosmetic products, maybe skin care, hair care, this sort of thing. And you develop these proteins, um, by very intentionally selecting amino acids, selecting their shape or structure, um, uh, I guess, rather, um, the shape and structure of the resulting protein, as well as then selecting the protein's, um, efficacy or expressibility.
Um, so help me here, does this sound like a good foundation for our conversation?
Lucile: That's an absolute good foundation. I think that was a really good, um, translation of what Cambrium is trying to do. We are a biotech company. We are [00:02:00] harnessing biology, uh, in order to, uh, design new proteins, in this case, ingredient for the personal care industry. AI is really in the center of what we are doing.
Uh, we'll talk about design and molecular design. I think, um, this is what makes us very, uh, different from the other players out there.
Deanna: Sure, sure. And I'm hoping that by learning more about what's going on at Cambrium, we can then have a better understanding of how AI is currently being used to support beauty ingredient development, um, more broadly as well. Let's think about how the process starts. If we were in Berlin with your team, what would be happening at the, at the protein design stage?
Lucile: Uh, the protein design stage is one of the most exciting for me. It's also, um, the one I really love talking about, uh, and why we use AI at that. stage to the end protein is because proteins are molecules which are programmable. Uh, if you think about proteins, like sequences of amino acids, which can be tailored into different functions and shapes, uh, you can also use the analogy of a necklace, a pearls necklace, right?
[00:03:00] Where every single amino acid would be a different pearl of different colors and shape, which give essentially in the end of the day, your proteins, different properties for the skin, for the hair. Let's say emollients. Let's say strengthening. Um, and we are able to use, uh, AI to, uh, tailor this design and enter into molecule, molecular designs thanks to AI.
Uh, so I did, uh, AI design stage. You would see a lot of people on the computer, uh, at Cambrium instead of being, uh, not necessarily in the lab. So in, at Cambrium engineers work together with scientists. We would see it as a computer with a material design brief, collecting all the ideas in terms of property that we want to have.
And, um, we would translate that into something that's, uh, our AI can, um, use in order to have the best prototypes of proteins that we would later put in the lab. Um, the good thing about AI here is that it has the power of looking into corners where the human [00:04:00] brain is not necessarily thinking about. So I see here.
Of course, it's more efficient to design with AI, but it also makes you look into places where you would not necessarily look at. And this translates directly into novel properties or unexpected, uh, benefits that you later have in your product. So that would be the first phase, uh, protein design, very computational, uh, scientists.
delivering a design brief, an engineer sitting in front of a computer, um, literally designing proteins in silico.
Deanna: Mm hmm, mm hmm. And you've told me that, um, as part of that process, then there's sort of a protein screening, um, operation that goes on as well. What's happening there?
Lucile: Um, so here it's, uh, maybe for people who've seen a lab before, it's maybe more conventional. So we are putting all our targets into the material world. So we go in the lab. We have a high throughput lab here in Berlin where we are able to screen. All the candidates that the computer told us are worth looking at, [00:05:00] that can be thousands of candidates, um, that we would look at, um, in our own lab.
Um, and then we are just looking at how the microbes from our biotech, um, platform, are they producing the sequence of interest? Uh, are they also responding the way we think they would in terms of material properties and we are selecting this way. So, uh, in terms of fields, but also in terms of efficacy, we are simply looking at, um.
What we design with the computer and trying to understand if the computer was actually right.
Deanna: Yeah, no, that's so, I hadn't really thought about, like, how right is it? Can you give us a sense
Lucile: It's, yeah, it is growing and growing. The more you work with it, the more correct it becomes, right? So when you start working, maybe you're 30 percent efficient into your predictions today. After four years of existence, we actually much more efficient. I think for certain part of the design, we were 70 percent correct, right?
That means that, um, yeah, we can really rely on our models now, um, to design [00:06:00] proteins.
Deanna: yeah, no, that's super cool. Um, I want to think a bit now, um, if we can, about, environmental sustainability, and this idea of digital twins, um, and I, I think that brings us to sort of fermentation and scale up in your process. Talk about, uh, that aspect of, of, of protein development design.
Lucile: Yeah, um, I'm a, I'm a chemist by training. So to me, precision fermentation was really something magic when I discovered it. So it's the power of harnessing biology in this case, microorganism to grow ingredient and chemicals as opposed to extract them. From, uh, different sources, uh, in our case, we use yeast, which are our microorganisms, which would convert carbon sources into proteins.
So you basically take a chemical input and you transform it into a protein. Um, this process, of course, decreasing the carbon footprint of our ingredients because we don't extract materials, but we really grow them. Um. With different carbon sources, there can be sugars and [00:07:00] other sources. Uh, I always make the analogy with, with beers, right?
Um, beers are precisely using the same process. Uh, you put carbon sources, which means sugar into yeast, which transform, um, the sugar into, uh, alcohols. So, um, that's essentially the same. And it's, of course, uh, more sustainable than, uh, what we had before. Our extractive methods, like, for example, taking collagen from bovine origin, even though bovine origin is extremely efficient, but the process is questionable.
And, uh, you mentioned digital twins. I really like this picture because you can really imagine, um, a digital twin of almost everything. Um, it's, you give us an input, the computer, um, all the characteristic of the twin that you want to build. So let's say I want to build, uh, a DNA digital twin. I would basically feed the computers with information which would tell them how you would react in certain situation.
And we do exactly the same, but with bioreactors. So bioreactors are [00:08:00] the machines which are hosting the precision fermentation. And when we are feeding the computer with information about how the microorganisms are reacting at this temperature, this pH, uh, when they are mixed at these speeds, we are able to predict their behavior and to take much more informed decision, uh, to scale up our process and become more sustainable.
Deanna: Yeah, no, that makes sense. And, um, just to help clarify, um, for myself and, and anyone listening who's, who's still learning along with me, um, I don't say this to trivialize the technology, but in my imagination, as you've said, right, you could create a twin of me, you could create a twin of a bioreactor, in some ways, a digital twin is kind of like a computer game or a software version of, in your case, a protein production facility, um, and I guess I'm curious, um, you've said a little bit about how you might program these twins, maybe you can say more there, but also, um, the AI functionalities that digital twins rely on, I think, um, [00:09:00] I think we're, we're getting, um.
In some ways, and I don't mean you and I, but I mean the conversations and articles I see around artificial intelligence are kind of becoming generic, like we all assume that now we know what generative AI is and that's how everything works, but there's a lot going on with artificial intelligence technology.
What sort of functionalities are supporting digital twins in your, in your
Lucile: Um, you, you mentioned, uh, generative AI, and I think that's a great way to use it. You're right. It became very common for people to use this word. It's the simple science of training an algorithm to react in certain ways, um, which are new for it. Right. So, um, in the case of chat GPT, it's relying on a lot of different information that it finds, uh, online, and then it's able to react and to create generate.
That's why we call it generative AI. a new answer based on the information it got before. In our case, we use generative AI to try to understand reactions that [00:10:00] we didn't study before based on a very large, large set of information. So it's essentially working the same, except that our use case is very different.
Uh, we look at biology. The health of ourselves, how to make them happy in the bioreactor to produce exactly what we want in a very efficient way.
Deanna: interesting. Yeah. Yeah. Thank you for that. Do you have an example? Um, you know, maybe something that the digital twin showed you, um, and then you made a change in your literal, your physical production facility.
Lucile: Yeah, for sure. There is two answers to that. The first reason is the efficiency of the process, which I think is a little bit of a boring answer,
Deanna: No, no, we love it.
Lucile: an ingredient manufacturer, it's, of course, always very important to look at your yields. And we are able, for example, for the digital twins of the bioreactors to estimate temperature or oxygenation rate to support ourselves to be happy and to produce more.
So this is typically things that we use every day at scale up and at production. I think a little [00:11:00] bit of a.
Deanna: let me jump in there like while you're looking at that. So you're looking at the digital twin and it's saying you would get a better yield today if it was half a degree warmer and you make that change and. Hopefully see that same result. Is that is that what you're suggesting?
Lucile: Exactly.
Deanna: Okay. Yeah, and you were going to share more
Lucile: Uh, I think where it becomes interesting is when AI is looking into corners you didn't think about. Um, so when we were designing our first vegan collagen sequence, um, we got a couple of them, um, within which we identified an antioxidant active peptides. Um, And first we were very skeptical.
We were like, what's the point of having a collagen, which is anti aging, but also antioxidants. This was actually a suggestion of the AI to have this active peptide specifically into the sequence that we would put to the material world. Um, we tested it and it appeared that this specific sequence was actually the most efficient as anti aging and also, uh, as a symbiosis effect in anti oxidation.
[00:12:00] We didn't think looking at antioxidants peptides when we designed the protein, but the AI suggested that it could be actually. Um, a pretty good idea to do so.
Deanna: Yeah. Thank you for that Um, so and I think what you've shared there is an example of how this digital technology can impact Beauty ingredient innovation in particular, but I'm hoping you'll share Sort of a more literal case study, I guess, because you do have, um, at least one ingredient in the marketplace that was developed with this technology.
Can you can you sort of share that as an example? So we can again, just continue to think about how AI is showing up in the beauty ingredient space
Lucile: Yes, for sure. So NovaCol is the first, um, vegan collagen that we design, uh, which came out of our platform as Cambrium's first product. It's an active ingredient for anti aging, um, the same way that you have other collagen or plant based other extracts. We designed it to be, uh, one micromolecular [00:13:00] and second skin identical.
We believe that the micromolecularity was very important for skin penetration to have effect in the dermis. Um, which we think collagen could not supply before because it used to be a very, very large molecule. And we also designed it to be skin identical, to be recognized by the skin as its own. That was kind of the initial brief that we translated into, let's say, more technical inputs to the, uh, to the computer.
Uh, to have different sequences. Uh, what appears is that we see that this specific molecule is actually supporting the whole collagen cycle. So we don't only boost the collagen synthesis as we could imagine, but we also supporting the collagen to assemble on itself to create more strong fibrils and density into the skin, which is also a very strong anti angic tool.
And we're also protecting the skin against its own breakdown. So we are. Slowing down the MMPs action, for example, which are usually having the, the role of chopping down your collagen as you're aging. And we also have this, um, [00:14:00] antioxidant properties that I talk about. So I think here in this use case, what we see is that we as a benchmark against the, the, the rest of what is available in the industry, of course, we have this collagen boosting ability.
We also have additional properties, um, that we design against, which open a totally new realm of, of, um, of properties for, for Novacol and, uh, collagen on its own.
Deanna: and have you created any sort of product prototypes with this ingredient or worked with it. You know, manufacturers or formulators who are working with it. Are there any unexpected challenges or, um, or, you know, how, how is it in formulation? Do you know that?
Lucile: In formulation, it's, it's absolutely stable. Um, it's easy to formulate, um, and very big advantage also of working with AI is that we don't input only parameters which are efficacy related, but also, um, formulation and safety, [00:15:00] um, data related, right? So we are able to say we want the peptide for a skin usage or skincare usage to be stable at pH four and five.
Uh, we also want it to be a skin identical for its biocompatibility to have also something which is safe. Um, so, so far, um, it's actually on the shelf with different brands, um, in the U S uh, and in Europe. Yeah, to summarize on this use case, we designed for efficacy, but it's also very interesting to have the ability to design for ease of formulation as well as safety.
Deanna: that's very interesting. Thank you for sharing that. And,
Before I ask my next question, I want to remind you that our industry's most important trade show opens on Thursday, March 20th in Bologna, Italy. At cosmoprof Worldwide Bologna, you will find more than 3000 exhibitors from 65 countries. You can explore all the companies and brands that we'll be exhibiting this year in the cosmoprof Digital Directory.
And cosmoprof Worldwide. Bologna, [00:16:00] if you don't already know, is dedicated to literally all sectors of the beauty industry. From the supply chain to finished products. There are conferences, educational sessions, special projects, and so much more going on at Cosmo Prof. I will be there in the Cosmo Pack section recording new episodes of the Cosmo Factory Podcast for you.
And there is a new exhibition layout this year, so you'll want to look at the map before you go. Tickets are only available online. You'll visit cosmo prof.com to get your ticket today. In the show notes to this episode, I will put, uh, the link for you to get your ticket to cosmoprof Worldwide Bologna. I will also share a link where you can explore the digital directory of exhibitors and you can find a link in the show notes also, uh, to a map of this year's new exhibition layout, cosmoprof Worldwide.
Bologna opens on Thursday, March 20th. I'll see you there. you reminded us there that, um, you are looking at efficacy, [00:17:00] particularly at the ingredient design stage. But I'm wondering, Um, the opportunities for using AI for ingredient efficacy testing sort of once it's developed or even formulation efficacy testing.
Can you, um, you know, it does it does it does it predict that end of real life for you as well? Are you using it in that way?
Lucile: This is also something we're using. We are better at using AI upstream of our process. So the design phase and the screening phase as well as the production. We start using AI also to support our design of experiments. So when we have a very large data sets and we want to, um, Enter into complex, um, design of efficacy testing.
Sometimes we can also use, um, certain techniques with loud data sets. They are not AI anymore, uh, which support us to have more chances to get the results that we want to see. Um, that's, uh, techniques that some laboratories are using already, um, but I think it's, it's also showing that, uh, you don't need [00:18:00] to use necessarily AI systematically to become better and data driven.
You can also use, uh, you know, classical statistical methods, uh, to support your work.
Deanna: That's an important reminder, right? This is 11 more helpful technology. It's not everything in the future.
Lucile: Mm hmm.
Deanna: I like that. So I guess my last question, you know, I, I ask something like this to a lot of folks that I, I chat with about, um, innovative technologies. Um, but I'm imagining, you know, the fact that these ingredients are designed using AI, might eventually be shared with the end consumer.
As you said, um, your ingredient is in products on the shelf. You know, are, are beauty lovers ready for AI ingredients? Is this something that there's a, um, a demand for?
Lucile: That's such a good question. Um, something to say about AI is that it's allowing us to develop ingredients much faster. So, you know, that ingredient discovery is quite a lengthy process. Um, and very often, unfortunately, in the [00:19:00] industry, we have to react. To consumer needs as opposed to lead with real innovations.
It's, it's much more complicated because putting an ingredient to in the material where it takes some time, I think the chance we have to take here is that is using AI to be more, um, dictating or leading the trends as opposed to try to react. And I think here I really see an opportunity, um, to use it also to.
Try to also, um, get to new novel properties that the consumer didn't even think about. Um, then are beauty lovers ready for that? That's, that's a very good question. I think there is always appetite and demand for very science, uh, backed ingredients, uh, which we are always trying to deliver. And I think the industry is so good at putting so many innovation every year on the market.
It becomes more and more challenging to actually come with new ideas and new ingredients. So I think there. I would welcome AI there as a tool to support us moving [00:20:00] faster with better ideas in the future, uh, to have consumers, um, getting to actually what, what they want, uh, with novelty.
Deanna: Yes. Interesting. Thank you so much. Well, it's clear from what you've shared today that AI supported molecular, uh, development is, is, you know, a significant step toward the industry's future. Lucille, I thank you for joining me here on the Cosmo Factory Podcast.
Lucile: Thank you so much for giving me a platform also to talk about AI and supporting, raising awareness about this topic, uh, in the chemical world.
Deanna: You're welcome. [00:21:00]