Eliminating excessive false positives is the holy grail in gene discovery – it saves money and time. Its Gene Discovery by Informationless Perturbation (GDIP) method has reduced false positive rates by 10-100x (depending on the trait and the organism) and they are bringing the power of GDIP to breeding climate-resilience traits into crops ranging from rice to rubber-producing plants. Avalo.ai is the seed company where sustainability and crop science are converging.
With a pinkish glow, the silvery door opened to reveal trays of dirt and seedlings. Held at a cool 54°F, Avalo CEO Brendan Collins gingerly removes the trays and proceeds to count with surgical concentration. Two new seedlings had popped and, notably, one their algorithm predicted would be the highest performing knock-out.
Based off a computational platform CSO Mariano Alvarez developed during his postdoctoral fellowship at Duke, Avalo’s ultra-fast, high sensitivity, and high accuracy gene discovery platform is being used to generate cold tolerant rice strains that can stave off the harvest shattering effects of weedy rice and late frosts.
Gene discovery itself isn’t new, but the gold standard, Genome Wide Association Studies or GWAS, is fraught with a major problem: too many false positives. A plant breeder who wants to target a specific trait is left with hundreds of gene targets to address, the vast majority of which are a waste of time (read: $20M dollars worth of time and effort wasted for each wrong gene target).
We’ve seen how this type of baked-in R&D cost is passed down to the payers in pharmaceuticals. Outside of major row crops, the agriculture industry simply cannot bear these types of excessive development costs. Enter Avalo’s Gene Discovery by Informationless Perturbation approach (GDIP, for short). Powered by Prof. Cynthia Rudin’s pioneering work in interpretable artificial intelligence, Avalo is using GDIP to solve the urgent problem of developing climate-resilient crops to feed the next billion people.
The excitement around Avalo’s early data is palatable. The 5 month old company has initiated four new academic and industry partnerships during IndieBio and last month it raised its Seed round. Brendan and Mariano are building their team and churning out new performance data every week.
We are creating the genetic blueprints for the fastest sustainable transformation in agricultural history.
Brendan Collins, CEO of Avalo
Yeah, totally!
Mariano Alvarez, CSO of Avalo
During SOSV’s IndieBio program we asked you prove your next generation gene discovery for the future of plants works. What did you discover?
We needed to show one major thing: that GDIP–Gene Discovery via Informationless Perturbation–finds usable gene targets the first time. We decided to take some publicly-available data on cold tolerance in rice (a major problem for California growers) and try to find a target that we could break quickly and inexpensively. At the Phytotron in Duke University, we were able to validate four genes for cold tolerance over a three month experiment for about $15,000. Pretty good compared to the alternative!
Can you explain a little bit about how your predictive algorithm actually works relative to current approaches?
We know that biology is inherently complex. However, to interpret biology, we have to simplify the model and that is what we are doing every time we run a genome wide association study. This allows scientists to bring some interpretability to biology, but the downside is that it lets in hundreds of false positives. By employing interpretable machine learning we can create a much more realistic model of biology that doesn’t compromise on complexity and using our informationless perturbations (our proprietary technique to explain black box models) we can probe that model and make discoveries.
Why does predicting crop or plant phenotype matter?
Plant products are all around us – not just as food, but as cosmetics, perfumes, and industrial feedstocks. The phenotypes of the plants – from the color of an aesthetically pleasing flower to the ability of a crop to shrug off infection – usually have some genetic component that we can amplify to make the plant more useful.
The problem is that most traits are the result of complex architectures, many genes interacting with each other and the environment to produce the observable outcome. This makes both gene discovery and trait prediction… well, hard! Current genetic tools cannot accommodate this complexity. We make it easy.
Who is Avalo?
When we started this whole thing, we were just two scientists from North Carolina with a good idea. Mariano was a postdoctoral researcher at Duke University with a background in plant evolutionary and computational biology. Brendan has a background in cell and clinical biology and has spent the last seven years doing software development for startups around Durham.
IndieBio really believed in us from a very early stage. Now, we’re growing to a team of 7 by the end of the year, with a team of superstar scientific advisors including Cynthia Rudin (head of the Prediction Analysis Lab at Duke) and Daniel Pomp (Professor Emeritus of Genetics at UNC Chapel Hill and co-founder of GeneSeek) as well as an amazing group of investors and supporters behind us. We feel pretty lucky.
We often joke that Avalo will develop partnerships on all 7 continents — from rice to wheat to dandelion to corals — do you think you will develop something for Antarctica?
That’s funny — our first in planta trial was cold tolerance in rice, so maybe we’ll just ratchet that up a bit. There are some cold tolerant varieties in Japan, so… maybe? In all seriousness, I think there are some really interesting opportunities to work on marine organisms in Antarctica – reefs, microbes, and even fish. We’d love to find a partner for those projects in the future!
