Avalo—a startup deploying interpretable artificial intelligence to develop climate resilient crops—has raised an $11 million Series A round and entered into a partnership with CCEP to futureproof its sugarcane supply chain and reduce scope 3 emissions.
The round was co-led by Germin8 Ventures and Alexandria Venture Investments with participation from Coca-Cola Europacific Partners (CCEP), Trust Ventures, Trailhead Capital, and angel investor Will Canine. Existing investors AtOne Ventures, Better Ventures, SOSV, and Climate Capital also participated.
The capital injection will help North Carolina-based Avalo to deploy its rapid evolution platform to bring low-input resilient crops to market, starting with cotton and sugarcane that can grow with less nitrogen fertilizer and less water.
“We’re very excited to be supporting Avalo’s expansion into the sugar category. Ingredients represent one of the hardest to abate emissions areas in our value chain and by tackling the problem at its source, Avalo has the potential to create more sustainable crops – reducing our scope 3 emissions and adding value to growers and other stakeholders across the value chain.” Nicola Tongue, CCEP Ventures
Climate resilient sugarcane
The partnership with CCEP will focus on sugarcane, a crop heavily reliant on water and fertilizer with limited growing regions, Avalo chief marketing officer Nick Schwanz told AgFunderNews. Complex genetics and lengthy breeding cycles in turn have made it difficult to improve through traditional breeding, he added.
“In Australia, runoff from nitrogen fertilizer is killing the Great Barrier Reef and Australia’s $2 billion dollar sugarcane industry is facing considerable headwinds. While drought and pest pressure continue to increase, the sugar being grown has remained largely unchanged for 15 years. This is where Avalo comes in. Not only could evolved sugar varieties [that can thrive with less water and synthetic fertilizer] future proof this at-risk industry for farmers and producers, they will also prevent further damage to critically endangered ecosystems.”
According to Schwanz, bringing a new sugarcane variety to market might typically take 12+ years. “Avalo could make it possible in just 5-6 years.”
Interpretable artificial intelligence
Avalo’s core technology—Gene Discovery by Informationless Perturbation (GDIP)—draws upon Professor Cynthia Rudin’s work in interpretable artificial intelligence to more rapidly identify the genes responsible for specific traits in crops, said CEO Brendan Collins, who cofounded Avalo with Dr. Mariano Alvarez in 2020.
“I think the larger blow up of AI has really muddied the waters. But one thing I really want to emphasize is the math that we are using is brand new. It was basically invented in 2018, whereas the math that goes into large language models was basically invented in the 1940s, we just didn’t have the computing power to pull it off until more recently. But what we are doing is bleeding edge on machine learning theory and AI and interpretability.”
He added: “I do think AI is going to have meaningful impact across agriculture, including the large language models, and things like computer vision for precision agriculture. But in genetics, there is not a huge amount of historical data to power those LLMs in a meaningful way, especially in crops beyond maize and soy. And that’s why we think our version of AI and ML is extremely valuable as we can make the most of the sparse and limited data that there is.”
Searching through a ‘sea of biological noise’
According to Alvarez, at the core of Avalo’s GDIP tech is “a way to identify important genetic information in a big sea of biological noise.”
This has two key applications, he said. “The first is identifying the genes that are responsible for a trait, and that could be disease resistance, heat resistance, or value-added traits for the downstream supply chain. The second is that it allows us to create models that are much more predictive of plant performance than we would have been able to create otherwise, using much less information.”
It quickly became apparent, he said, that “using that predictive component would allow us to run a breeding program, a genomic selection program, at much lower cost with much higher efficiency.”
Historically, the main approach to connect genes with traits is the GWAS, or genome wide association study, he said. By scanning the plant genome for markers that consistently appear in plants with a desired trait, breeders can pinpoint regions likely responsible for that trait, and use that data to guide breeding decisions.
“You’re basically going gene by gene or allele by allele, and saying, ‘Is this allele associated with this trait of interest?’” said Alvarez. “And then you do that 100,000 times or 500,000 times or a million times. If you have a trait caused by a single allele, you can get a very clear and accurate picture of that trait.”
The challenge, he said, is that most of the traits we care about in agriculture are not single gene traits. “They’re polygenic traits, which may have hundreds or thousands of genes underlying them, and in that situation, it’s very difficult to determine which genes are important because they’re all working together and interacting.
“Our model, rather than looking allele by allele through the genome, uses a machine learning model to look at the entire genome all at once, all at the same time. Rather than asking, ‘Is this allele associated with this trait?’ it asks, ‘Is this more associated with the trait than all of the other alleles that I’m looking at?’
“And is this conditionally more important, rather than just individually important? And that allows us to develop a much better baseline for how the genome is working and more accurately assess which places in the genome might play an outsized role compared to everything else.”
Dramatically speeding up traditional breeding
According to Alvarez, “With GWAS, maybe 15% or 20% of the things you identify are truly causal of the trait you’re looking at. Our model is greater than 90% accurate.” To test this, he said, “Greater than 90% of the things that we find when we knock them [the offending genes] out in a plant are changing our trait of interest.”
At base, he claimed, “The things that we find are then much more likely to be validated in a practical setting.”
Collins added: “Traits such as drought resilience and nitrogen efficiency can be driven by more than 1,000 alleles across the genome. By being able, with really high fidelity, to hone in on these allows us to take advantage of all the natural diversity that has been collected and in seed banks globally.”
The business model
So once the genes of interest have been identified, what happens next?
According to Alvarez, Avalo’s initial focus is using its predictive models to accelerate traditional breeding programs, although he noted that the approach could also inform gene editing applications.
“It’s trait dependent. If you’re looking at pests, they’re usually exploiting a couple mechanisms within a plant, so maybe it’s only a couple genes you would need to edit to increase pest resistance. But there’s not a single drought gene or one yield gene. So with current gene editing technology, even if you could make the number of edits needed to make to make a substantial impact on those traits, you’re going to change so much else about the plant itself.”
The business model will vary depending on the project, meanwhile, said Collins. “In the sugarcane model, we are developing low nitrogen, low water sugarcane for Coca Cola to reduce their scope three emissions, and they are willing to pay a premium for that. That allows us to go to farmers and say we can contract grow with you to get this commodity out to this big CPG and you just need to follow this agronomy recipe and use our seeds.
“That gives us evidence to go to investors and say, we need this amount of money to develop this crop. We have this offtake guaranteed. And then by reducing the inputs that have to go into the field, we can increase sustainability and farmer profitability and get some margin for ourselves.”
In this scenario, he said, “We’re basically doing toll-based manufacturing. We’re giving farmers seed at a discounted rate, helping them with agronomy to grow our seed in their environments, and then harvesting with the farmer and selling the sugar at the post farm gate into the commodity supply chain, so we’re sharing the profits from their sugar, rather than selling farmers sugarcane seeds.”
Next steps in CCEP sugarcane partnership
As for CCEP, said Collins, “We’re not going to crack this in a year, and this is a multiyear project, but it can be faster than it was before, and way less expensive.”
Sugarcane, he noted, is a tropical crop. “So by being able to change its water tolerances, we can also change the regions it’s grown in and put it into lower impact environments [that don’t contribute to deforestation for example].”
Avalo’s goal is to have its “first machine learning models trained this year so we can start making recommendations at the end of this year or at the beginning of next year,” said Alvarez. “We really want to make an impact in the supply chain in the next three to five years.”
Broccoli, rice, rubber, cotton…
Other areas of focus beyond sugar are staple crops such as cotton, rubber, and rice, said Collins. “Texas is by far the largest cotton growing state in the United States and they have experienced unbelievable droughts the last couple years, but the industry has been breeding varieties that perform in high irrigation and high nitrogen environments.
“We are targeting the exact opposite to create low-nitrogen use cotton that can be completely rain fed, but still have premium fiber characteristics, and our progress so far has been pretty awesome. We are going to be on 2,000 acres this year. We do have a partner for cotton but we can’t announce it yet.”
Avalo is also working with partners to create a US source of rubber production through the latex produced in the roots of particular species of dandelion, he said. “It’s weird to think about rubber crops being grown in Ohio, but we’re hoping to release a variety in the next year or two.”
Other projects include fast-growing broccoli, enabling farmers to get six, rather than four harvests a year from an indoor facility. Aside from the obvious economic benefits of more harvests per year, growers can also stop using pesticides as the crop grows too rapidly for pests to gain traction, said Collins.
“There’s no pest life cycle that is shorter than 45 days. We were able to get a crop that was harvestable in 37 days.”
Fundraising
Asked how challenging it has been to raise money in the current environment, Collins said: “People are generically skeptical of agtech right now, so they are looking for market signals from corporates [that there is commercial interest in their tech]. So having CCEP on board gave a strong market signal.”
Michael Lavin, managing partner at Germin8 Ventures, which co-led Avalo’s Series A round, told us: “Agriculture is a national security issue inextricably linked with America’s ability to be food secure, economically productive throughout the supply chain, healthy and competitive internationally.
“With a powerful AI capable of harnessing the causal genes for breeding complex traits on the fastest timelines—reducing what is normally accomplished over decades to just a short few years or less—Avalo is liberating the most needed genes that already exist in the plant. This will help commercialize varieties which break barriers on the farm and are purpose-designed for society’s most pressing commercial needs, at the pace of business.
“While Avalo’s AI yields an asymmetric advantage, make no mistake, this is a market-maker and product-first enterprise leveraging an unmatched ability to empower and transform supply chains from farm to manufacturer.”
The post Exclusive: AI-powered plant breeding startup Avalo raises $11m, partners with Coca Cola to future proof sugarcane production appeared first on AgFunderNews.
