On this sweltering West Texas afternoon, you’re looking for genes that will give future generations of cotton plants a better chance of surviving this hot, dry, changing climate. No problem – if you’ve got a couple billion dollars and 15 years to plant, observe, harvest, sequence, then do it all again. That’s the way Big Ag does it. But AI crop development company Avalo.Ai has found a better way.
Using a variety of AI-based tools, from advanced computer vision to proprietary Interpretable Machine Learning technology spun out of Duke, Avalo’s small team turns to nature’s own wisdom to find hardy code hidden in a crop’s wild ancestors. Then they use their AI foundation model “Gaius” to run simulations of 10,000 years of evolution in seconds. The result is a highly cost effective way to identify nature’s best solutions for resilience — and any other desirable trait — and breed them into pretty much any crop they want. All within just one or two breeding cycles.
They’ve already done it for cotton in Texas, where the industry is nearing collapse due to climate change. And that’s just the beginning. But as with most deep tech companies, Avalo’s path to this incredible threshold was anything but linear.
Rewind a decade and a half. Avalo CEO Brendan Collins was pursuing a Ph.D. in clinical neuroscience in London, working on cell-based therapies for traumatic spinal cord injury.
He loved research, but grew disillusioned with academia, so he left. Moved to North Carolina. Founded a startup that would pay student athletes to graduate. Got 256 cease-and-desist letters from the NCAA. Closed up shop. Built software for a robotics company.
Meanwhile, Avalo CSO Mariano Alvarez was happily ensconced in academia with his eyes firmly on the prize of a professorship. After a Ph.D. in Ecology and Evolutionary Biology, Dr. Alvarez was trying to create a computer model that factored in climate and genetic variables to predict whether a plant would be able to survive climate change. But he ran into a wall-sized issue: deciphering the genetic basis of complex traits is insanely challenging.
Mariano and Brendan’s paths converged, serendipitously, at a brewery. Over beers and brainy banter, they soon became best friends. With his academic background and love of research, Brendan grokked Mariano’s work – and its challenges.
Traditionally, if you want to understand the genetic basis of a trait – say, how tolerant to cold a variety of rice is – you use GWAS (a Genome-Wide Association Study). Compare the DNA of thousands of rice plants, marker-by-marker, and see what the hardiest survivors have in common.
Problem #1: False positives. Since genomes are staggeringly similar, this approach gets tricked by unrelated genes that different plants coincidentally have in common.
Problem #2: proving if a gene actually enhances cold tolerance (or any other trait) requires physically altering and growing the plant in a lab. This costs time (3-6 months) and money (50 to 500k!) for each gene. And if it’s a dud? That’s time and money flushed down the proverbial toilet.
And our agricultural system doesn’t have the time for that! So, Mariano walked into the Duke University lab of professor Cynthia Rudin, a prominent computer scientist and pioneer in the field of interpretable machine learning. (or as Brendan describes her, a “God-tier statistician.”) Collins explains, “Together they created this framework for applying interpretable machine learning to the genome for the first time. And it really solved the false positive problem.”
Enter: “G-DIP” – Gene Discovery via Informationless Perturbations. Essentially, instead of having to look at each and every marker in a genome one by one to find a tiny trait needle in a massive genomic haystack, G-DIP allows you to look at all the positions simultaneously. And Avalo’s proprietary tech lets you ignore the hay (aka the false positives and irrelevant markers). G-DIP zeroes in on the good stuff (the best gene candidates for the trait you’re interested in). This reduces the scale of your homework assignment from looking at millions of markers down to several thousand markers.
There’s more to it than that (hint: drones, decaploid genomes, lasers, dirt under your fingernails, etc.) but first, back to the brewery…
Collins was a good practical coder and offered to use his experience in building software to supercharge Alvarez and Rudin’s G-DIP technology to production scale. “Something that could reliably run on huge data sets.” And he tried to convince his drinking buddy that they could use this tech in the “real world.” But Alvarez was set on becoming a professor.
“I went on a hardcore lobbying campaign to say that maybe academia isn’t all it’s cracked up to be, and it’d be more fun to do this as a startup.”
Brendan Collins, founder & CEO, Avalo.ai
So, as Collins explains, “I went on a hardcore lobbying campaign to say that maybe academia isn’t all it’s cracked up to be, and it’d be more fun to do this as a startup.” And they made a bet. If Collins could go raise some money, Mariano would leave his ivory tower and they’d do this thing as a startup full time.
Enter: SOSV
Brendan reached out to a variety of potential investors, but found SOSV particularly appealing. “I don’t think there was any other early-stage backer that had had the wins that SOSV SF had, and was so focused on deep tech.” He applied and they were accepted in early 2021. Bet won, Brendan called in his chit. Mariano took off his mortarboard and they both rolled up their sleeves.
Over the next couple months at SOSV SF, they used their tech on an existing rice data set. Could G-DIP find genes that would improve cold tolerance during the critical germination stage? If so, it would be a hell of a proof of concept. Rice seedlings are incredibly fragile, and as weather patterns become more erratic due to climate change, it gets harder for farmers to predict when to plant their crops. A late-spring frost could destroy a fledgling crop and, quite possibly, scupper the entire season.
“This rice data set had been analyzed a thousand different ways by people before us. We were able to find things in the genome that no one else was able to.”
Brendan Collins, founder & CEO, Avalo.ai
And/but, as Collins says, “This rice data set had been analyzed a thousand different ways by people before us.” In other words, this haystack had been thoroughly searched for needles.
But not with G-DIP. “We were able to find things in the genome that no one else was able to,” says Collins. Specifically – Avalo found five genes that helped improve cold tolerance during the germination stage.” And they were able to prove it in the lab while at SOSV SF. “We were able to actually do a [gene] knockout and show phenotypically that our algorithm validated every single time.”
All this within about four months. Minds blown. Seed round raised. Specifically, Avalo raised $2.9M two days before Demo Day. “It was a very relaxing demo day,” recalls Collins fondly.
But when they graduated from SOSV SF in July, 2021, they were still “very much like a cool technology looking for a market,” admits Collins. First, they tried to be a gene discovery company for industry. Their product would be a list of highly validated “true positive” gene targets. Big seed companies could buy that info and use it to create an optimized plant via either gene-editing (GMO) or a massive traditional breeding program.
That was the plan. But Collins and Alvarez soon realized that, “The big seed companies only cared about two things – herbicide resistance and pesticides resistance.” They weren’t buying what Avalo was selling. Collins and Alvarez had to pivot or perish.
They knew their AI was incredibly accurate at predicting natural breeding outcomes. And they’d proven it by growing cold-tolerant rice strains in the lab while at SOSV. What if they just grew the perfect plant themselves? Could they show the world they could achieve, “meaningful outcomes in an unbelievably short period of time?”
To do it, they needed three things: #1 A partner, #2 A Target Crop and #3 Data (lots of it!)
They found the perfect partner in a robotic/vertical farming startup called Iron Ox. Iron Ox had the necessary company values, infrastructure, and facilities to grow the vast amounts of seedlings they would need for a massive diversity trial. Together they decided broccoli was the perfect candidate – a valuable crop, but one that no vertical farms or greenhouses could grow commercially. The big headed Italian broccoli you get at the supermarket takes an eternity to grow (120 days vs. 30 days for something like lettuce) and other varieties, like broccolini, are notoriously difficult to grow in greenhouses and vertical farm spaces because they’re so heat sensitive.
In order to gather the vast amounts of data they wanted, Avalo planted over 500 varieties of broccoli, including “normal varieties,” plus wild predecessors and “edge cases” from seed banks to find rare, fast-growing and hardy traits that had been bred out of supermarket broccoli. Then, they set out to learn all that was learnable. As Collins explains, “We measured everything about these plants, including nutritional content, flavor, growing rates” and more. They even created 3D models of the plants to help measure head size and other favorable physical characteristics of the plants.
After feeding all that data into their AI, it identified the desired traits and ‘gene modules’ responsible for fast heading and heat tolerance. It also provided a genetic map and breeding plan for the team to follow. This allowed them to find a winning broccoli variety in just two breeding cycles.
And what a broccoli. Collins explains proudly, “We were able to develop a broccoli that headed in thirty seven days, and could tolerate the variable heat conditions that you would have in a greenhouse.” Thirty seven days. Farmers could grow three crops in less time than one traditional crop. But this fast timing isn’t just about speed of production, it’s also a sustainability play. Because many broccoli-munching critters have a life-cycle of 45 days, being able to harvest it in just 37 helps manage pests and eliminate the need for pesticides. A strategic win/win for farmers’ bottom line, and their soil health.
Sadly, Iron Ox became a casualty of downsizing in the vertical farming and greenhouse space, but the super broccoli collab was a great proof of concept of Avalo’s tech. (and it’s on the menu at chef Dan Barber’s restaurant Blue Hill in New York!) It also helped them secure Series A funding from a syndicate of VCs including Germin8, Alexandria, AtOne, Better Ventures, SOSV, and strategic partner Coca Cola Euro Pacific Partners (CCEP).
Flush with funds and fired up by their super broccoli success, Avalo started building a new business model – using their “nature-based, AI-powered, Rapid Evolution Platform™” to actually create new products. And they committed to, “Tackling the biggest externalities in broad acreage crops globally.” Specifically, the negative environmental and social costs seemingly inherent in many of the most critical crops the world depends on.
According to Collins, cotton uses more pesticides per acre than any other row crop on the planet. Rice produces more methane than any crop on the planet. Globally, sugarcane displaces huge amounts of biodiversity. And rubber causes an outsized proportion of deforestation. Avalo set their sights on these vital commodities, which they dubbed, “The Formidable Four™.”
If they could make an impact with them, it would bring farmers higher profits and more success – quite possibly making the difference between a farmer surviving or going belly up. They realized they could provide more value to brands and CPGs too, since their AI and data acquisition tools allowed them to optimize seeds not only for yield (key for farmers) but also for winning traits all along the value chain.
In cotton, for instance, Avalo could also identify and amplify traits related to fiber strength and length, which are key for spinners and weavers. These mid-stream players subject cotton to a grueling multi-step process in which cotton touches a physical saw blade eleven different times before becoming apparel. Traditionally this means less than a third of the cotton that leaves the field makes its way into your T-shirt. Stronger fibers means less cotton gets lost.
And by breeding plants that naturally thrive in dry conditions and require far less fertilizer, they could also help brands and retailers hit Scope 3 emissions targets they’ve committed to. For all these reasons, cotton rose to the top of their target list. And Collins and Alvarez started spending a lot of time in Texas.
With 8 million acres planted annually, Texas is the largest cotton-growing state in the US. Traditionally 70% of the cotton there (specifically in the Texas High Plains) has been irrigated, and cotton has been bred to thrive in a wet, nitrogen-rich environment. But due to aquifer depletion, now only 30% of cotton-growing land in West Texas can be irrigated, and those same cotton cultivars just can’t cut it.
Over the last several years, there’s been field abandonment rates of up to 70%, according to Collins. That’s when farmers decide to just walk away from a crop part way through the season because it’s doing so poorly it won’t be worth the money or effort to keep growing it. This has led to huge insurance payouts and existential challenges for farmers just trying to make a living. Collins sums it up, “There’s just this massive need on the ground for better genetics that farmers can use to actually make their livelihood.” Challenge accepted.
Enter: GAIUS
To fill that need, Avalo would once again turn to nature for potential solutions. And they’d use their growing AI stack to tease out the winning combination of genes from millions of years of genomic data. As they built more and more AI models or “experts” to oversee various areas of their operations, they also developed a tool they dubbed Gaius – a general purpose, decision-making, “foundation model” AI that presides over all the other AI “experts” and helps inform every aspect of Avalo’s business.
In 2024, they planted thousands of seeds from over 500 different varieties of cotton. They included current commercial varieties, wild progenitors and landraces (varieties developed over the centuries by farmers) from all over the globe. They put all these into the harsh, un-irrigated Texas soil and set about learning everything they possibly could – just as they’d done with broccoli. And this time, they employed even more cool tech and more AI controlling software and hardware to really focus their efforts.
AI-powered drones used advanced computer vision to scan the fields from the sky, inferring plant genetics visually. Out of the 150,000 individual cotton shrubs they’d planted, this enabled them to zero in on the most promising 15,000 plants, and harvest only those. Next, a robotic, hyperspectral sorter looked inside each seed as they whizzed by at 20 seeds per second, intelligently selecting the absolute best 2,000 out of over four million candidates.
Then, rather than having to sequence the whole genome, Gaius knew to ignore 97% of the data and told Avalo which 44,000 markers actually matter. This slashed sequencing costs – down to $17/seed from $60. And they just had to sequence those 2,000 MVPs.
Doing all the math, that means Avalo paid just $34,000 for a treasure trove of highly valuable data which Gaius then used to determine what cotton varieties should be bred together in the next growing cycle to get the coveted combo of drought resilience, fiber length, and other desirable traits. For a crop like cotton that’s planted once a year, Collins says, “We can do it from start to commercialization in three years.” And, he emphasizes, “We’re working on a suite of traits at any given time” – not just a single trait.
Putting this all into context, for the price of a well-equipped sedan, Avalo accomplished what would traditionally cost big ag $2.7 billion dollars and 10-15 years to achieve. For a crop with a single improved trait. And potentially “improved” by only 1%, which is the typical genetic gain a conventional breeding program delivers, according to Collin.
GAIUS enables “essentially 500 years of evolution in one year.”
Po Bronson, General Partner & Managing Director, SOSV SF
One percent. The brand new, climate-resilient cotton variety Avalo created for the Texas High Plains? It led to, on average, 111% higher yield, while maintaining fiber quality, all with zero need for irrigation. And it’s commercially available today. Gaius enables, “Essentially 500 years of evolution in one year,” marvels Bronson. That’s huge. As is the global cotton business ($50 Billion).
The folks who buy cotton are taking notice. After their first commercial year, Avalo has already struck up deals with massive global brands (which are still top-secret) looking to address their scope 3 emissions, derisk their supply chains, and meet the rising demand for more natural and sustainable goods.
And cotton is just one of the Formidable Four. Collins says, “Sugar, which we are reinventing with Coca Cola, is a $70 billion business. Rice, which we’ve worked on, is a $300 billion business… Natural rubber is a $50 billion market. We’re doing all of them.”
Quite a wild ride so far. From two pals at a brewery in North Carolina to a team of 19 tackling existential problems around the world, Avalo.ai’s path thus far may not have been predictable or linear. But their trajectory sure seems nothing short of stratospheric – with the potential to reshape the very roots of global agriculture.
[More information on the scale and scope of Avalo’s AI can be found in this great conversation between Collins, Alvarez, and SOSV General Partner Po Bronson.]