Before joining the company in 2022, he was director of metabolic engineering at Provivi, a biotechnology company that helps farmers protect crops with naturally effective and sustainable solutions.
Glenn’s academic research into the chemistry of plants is extensive, including his Ph.D. work at the Massachusetts Institute of Technology where he studied how the Madagascar periwinkle makes two notable anti-cancer agents.
“I’ve always been fascinated with plants and interested in the chemistry that they do,” Glenn said. “It was very fortuitous that I was able to get a position at Ayana Bio that really took advantage of the skills that I had previously in metabolism and plant metabolism in particular and in working with big data for the purpose of making molecules that are beneficial in some way.”
Working with big data means harnessing the power of AI. In this interview with NutraIngredients-USA, Glenn discusses the promise of artificial intelligence in helping the development of ingredients and where that technology may be headed for Ayana Bio.
How do you use AI at Ayana Bio?
At Ayana Bio, we use AI to understand how our cell cultures respond to different conditions so that we can get the cultures to make specific molecules.
We get a lot of information from AI. We examine the transcriptomics, meaning we look at which genes are being expressed under which conditions, and we look at the metabolomics, meaning we explore which molecules are made under these different conditions. Multiplying the hundreds of conditions we test by the hundreds of thousands of transcripts and thousands of metabolites we monitor in each sample gets us billions of features. These features become part of a big data set that you can use to predict what needs to be true for your cell lines to perform a certain way.
I think one of the big differentiators at Ayana Bio is that we use our AI to tell us how we develop one cell line, such as sage, and get it to make high levels of rosmarinic acid for a health benefit. And we can use the same cell line and get it to make high concentrations of triterpenoids, which have a completely different set of health benefits. AI helps us predict and control our cell lines with great specificity.
People are constantly talking about bioavailability in the supplement industry with the assumption being that if you increase the potency of a particular compound, that it will lead to more efficacy in the body. But is that always the case?
I want to be careful not to speak in generalizations, but I would say I don’t think it’s always the case that increasing the concentration of a particular molecule in a supplement always leads to increased potency because what matters is how your body responds to what you’re taking in. We know that oftentimes these molecules are transformed in the body, and that’s what actually has the health benefit rather than the molecule that you’re ingesting itself. We want to be sure that the ingredients that we’re making have the molecules that we want and that the molecules are getting to where they need to go in the required form to have an effect.
The way that people test bioavailability is by running preclinical studies using mice as an example. Sometimes people can use different model organisms like C. elegans. I can say that we are even planning out some potential experiments looking at bioavailability in a model organism like fruit fly. These models let you do some really cool experiments to see if the molecules that you’re interested in are actually getting into the brain of the organism. But I can’t say that we are using AI to look at bioavailability ourselves at Ayana Bio.
And how do we know which molecule is desired, and how does AI help facilitate that?
At Ayana Bio, we are focused on figuring out how to make bioactives in a plant cell line. We’re using our AI to produce bioactives at levels that have known health benefits. Other ways to use AI in the field more broadly include finding new compounds of interest that may confer a health benefit and determining effective doses of bioactives.
When people think of the pitfalls of AI, they often think of the hallucinations it might create. We see this as a problem when consumers use ChatGPT. But do you find that the AI tools that you depend on have this issue?
I want to be careful here. I know that hallucinations can happen with any AI tool that you’re using. So you need domain experts to validate what you’re seeing. We have our data sets validated or grounded. We take what we’ve learned from those predictions then go back and test them. I would say that we’re not as subject to hallucinations as people who are just using large language models, especially ones that are not as heavily trained. But it is always important to think about how you might be wrong. I also think that it’s very important to have good, high-quality data first and foremost. It is imperative to ground that data with domain experts. It’s always important to validate your results.
Considering these large language models for ChatGPT that scour the entirety of the internet versus the smaller subset of data you work with, do you think a smaller subset contributes to the accuracy of Ayana Bio’s information?
Absolutely. The data that we have is highly controlled. We’re not using these large language models that scan the entire internet. There’s a lot of good stuff out there on the internet. There’s also a lot of garbage out there that hasn’t been validated. As the adage goes, garbage in, garbage out, or better yet, quality in, quality out. We have real high-quality data that we’re able to draw on, so we’re not as subject to hallucinations.
What can you tell me about the use of AI at your company and what that means for potential supplements being consumed?
We want to be able to say that our products are highly standardized, consistent and effective. That’s what we want to be able to say as a company. Before I finish my answer, I should say that Ayana Bio operates in a business-to-business model. We’re not interfacing with the customers who would be buying supplements off the shelf. Our customers are other ingredient companies who are looking to either shore up their supplies of specific ingredients or consumer packaged goods companies who are looking to formulate our botanical ingredients into their products for specific health benefits. At Ayana Bio, we want to be able to say that our ingredients are consistent and that we have high quality products driven by our AI tools, by our machine learning.
We’re running programs with Ginkgo Bioworks, which means we’re getting more data. And we’re trying to standardize these data sets so that we can make them interoperable, meaning that you can compare plant No. 1 to plant No. 2 to plant No. 3 and so on. We want to be able to make comparisons between these big data sets and also future proof them in a way that lets us be as information rich as possible. The goal of having these data sets is to be able to predict and control. You want to be able to predict which conditions are going to lead to specific bioactives. You want to be able to control those processes so that you are making an ingredient that will provide a specific health benefit.
What does the future look like for AI and the work at Ayana Bio?
For the future of AI in our company, we are again going to collect large data sets that are interoperable. The future of AI is going to be looking at these data sets and not necessarily having to screen so many different conditions to predict which ones are going to lead to those higher concentrations of health-beneficial bioactives.
