Avalo uses machine learning to accelerate the adaptation of crops to climate change – TechCrunch

The effects of climate change are affecting agriculture all around the globe. There are many solutions, but they are not always easy to find. If you could grow crops that can withstand heat, cold, and drought, instead of having to move a thousand miles away from your farm, would you? Avalo makes it possible to grow plants that are resistant to heat, cold and drought using AI-powered genome analysis. This can help reduce the amount of time and money required for breeding hardier plants in this century.
Avalo was founded by two friends who decided to try their hand at starting a startup before they settled down to academic life. However, it has a direct value proposition that requires some science to comprehend.

The big seed and agriculture companies invest a lot in creating improved versions of the major crops. They can make corn and rice more resistant to heat, insects drought, flooding or heat.

Co-founder and CEO Brendan Collins said that there are significant drops in yields in the equatorial regions. However, it is not that corn kernels have been getting smaller. Salt water intrusion has caused farmers to move upland, but early spring frosts can kill their seedlings. They also need resistant wheat to withstand fungal outbreaks during humid, wet seasons. If we are to adapt to the new environment, we need to develop new varieties.

Researchers emphasize the existing traits of the plant to make these improvements. This is not about adding a new gene, but rather bringing out those already present. It was simple to grow several plants and compare them before you planted the seeds of the one that best represents the trait, such as Mendel in Genetics 101.

Today, however, the genomes of these plants have been sequenced and it is possible to be more precise. It is possible to target future generations by identifying the genes that are active in plants with desired traits. This is because it takes as much time as a decade to do this.

The difficulty in modern life stems from the fact that traits such as survival in the face a drought are not just one gene. It could be a combination of multiple genes interconnected in complex ways. There is no one gene that can make you an Olympic gymnast. There is no single gene that can make you drought-resistant rice. Companies conduct what is called genome-wide association studies. They then have to test a variety of combinations in live plants. This can take years, even at industrial scales and rates.

Mariano Alvarez (co-founder and CSO at Avalo), said that it is difficult to find genes and do anything with them. It is simple to improve the efficiency of an enzyme. You just need to use CRISPR to edit it. However, increasing corn yield can be difficult because there are many genes that could contribute to this. It's a lengthy process if you are a strategic [e.g. Monsanto] who is trying to create drought-tolerant rice.

Avalo is here to help. Avalo has created a model to simulate the effects of changes in a plant's genome. This can be used to reduce the 15-year lead time and increase the cost.

Collins explained that the idea was to create an evolutionary-aware model of the genome. This is a system that models both the genome and its genes with more context from biology as well as evolution. A better model will produce fewer false positives for genes associated with a trait because it excludes noise, unrelated genes, and minor contributors.

One company was developing a cold-tolerant strain of rice. He shared the example. The genome-wide association study revealed 566 genes of potential interest. Each investigation costs around $40,000 due to time, materials, and staff. This means that the cost of investigating one trait could run to $20 million over many years. Naturally, this limits both the parties involved and the crops they are willing to invest their time and money. You can't invest that much money in a niche crop to improve it for an outlier market if you don't expect a return.

Collins stated that we are here to make this process more accessible. The same data set was used to identify 32 genes of interest in cold-tolerant rice. Based on simulations and retrospective studies, all of these are causal. We were able grow 10 knockouts in three months to validate them.

Let's unpack the terminology. The Avalos system eliminated more than 90% genes that would have to be individually examined. These 32 genes were found to be not only related but also causally affected the trait. This was confirmed by short knockout studies in which a specific gene is blocked and its effect on the trait is studied. Avalo refers to its method as gene discovery via informationless perturbations (GDIP).

It is partly due to the inherent ability of machine learning algorithms to pull signal out of noise. Collins however noted that they had to approach the problem differently, letting the model discover the relationships and structures on its own. They also wanted the model to be explicable. This means that it should not appear as a black box, but has some explanation.

This is the most difficult issue. However, they managed to solve it by swapping out the genes of interest in multiple simulations with what amounts to dummy versions. These don't disrupt the trait, but help the model learn about each gene.

Our tech allows us to create a minimal predictive breeding set that identifies traits of interest. Collins said that you can create the perfect genotype in silica [i.e. in simulation], and then do intensive breeding to watch for that genotype. It is possible to do it with smaller groups or with less-popular crops. Or for traits that are not available because of climate change. Who can predict whether wheat that is more heat- or cold-tolerant in 20 years?

Alvarez stated that by reducing the capital costs of this exercise, we can unlock this space to be able to work on climate-tolerant traits economically.

Avalo has partnered with universities to help accelerate the creation resilient and sustainable plants. These research groups are able to provide a demonstration of the company's capabilities because they have a lot of data and not as many resources.

These university partnerships will prove that the system works even for plants that are not yet indomesticated and need to be improved before they can be scaled up. It might be more beneficial to grow a wild grain that is naturally resistant to drought than to try to increase drought resistance in a large species of grain. However, no one was willing spend $20 million to do so.

They plan to market the data handling service as a first-ever startup. This is one of many startups that offer big time and cost savings to established companies in areas like agriculture and pharmaceuticals. Avalo may be able, with luck, to bring some of these plants into the agriculture sector and also become a seed provider.

Just weeks ago, the company emerged from IndieBio accelerator and has already received $3 million in seed financing to allow them to continue their work at a larger scale. Better Ventures and Giant Ventures co-led the round, with At One Ventures Climate Capital, David Rowan, and IndieBio parent SOSV also participating.

Alvarez said Brendan convinced him that starting a startup was more exciting than applying for faculty positions. He was absolutely right.