Avalo: Accelerating Crop Adaptation with Machine Learning

The Future of Farming: AI-Powered Crop Resilience with Avalo

Global climate change presents significant challenges to agriculture, and effective solutions are often complex. However, the ability to cultivate crops inherently resistant to extreme temperatures or drought conditions offers a compelling alternative to costly relocation efforts.

Avalo is pioneering this future by leveraging AI-driven genome analysis to accelerate the breeding of hardier plant varieties, crucial for navigating the uncertainties of a warming world.

The Traditional Challenges of Crop Improvement

Large agricultural companies invest heavily in enhancing major crops. Even minor improvements in resistance to heat, pests, drought, or flooding can substantially boost yields and profitability for farmers.

“Yield declines are particularly pronounced in equatorial regions – not because kernel size diminishes,” explains Brendan Collins, co-founder and CEO of Avalo. “Farmers are forced to relocate due to saltwater intrusion, only to face early spring frosts that destroy seedlings. Or they require rust-resistant wheat to withstand fungal outbreaks in humid summers. Adapting to this new reality demands the creation of novel varieties.”

Historically, researchers have focused on amplifying existing traits within plants, rather than introducing foreign genes. This process, reminiscent of Mendel’s foundational genetics experiments, involved growing numerous plants, comparing their characteristics, and propagating those exhibiting the desired trait.

Modern Genomics and its Limitations

Today, with plant genomes fully sequenced, a more targeted approach is possible. Identifying genes active in plants with desirable traits allows for enhanced expression of those genes in subsequent generations.

However, even with these advancements, the process remains time-consuming, often spanning a decade or more.

A key difficulty lies in the fact that complex traits, such as drought resistance, aren’t governed by single genes. They result from the intricate interplay of numerous genes. Consequently, genome-wide association studies often yield hundreds of potential gene candidates, requiring extensive and laborious testing in living plants – a process that takes years even at an industrial scale.

Numbered, genetically differentiated rice plants undergoing testing. Image Credits: Avalo

Avalo’s Innovative Approach: Predictive Genome Modeling

“Simply identifying genes isn’t enough when dealing with complex traits,” states Mariano Alvarez, Avalo’s co-founder and CSO. “Enhancing enzyme efficiency through CRISPR is relatively straightforward, but increasing corn yield involves thousands, potentially millions, of contributing genes. For a major company aiming for drought-tolerant rice, the investment can reach $200 million over 15 years – a substantial undertaking.”

Avalo addresses this challenge with a proprietary model that simulates the effects of genomic changes, drastically reducing both the time and cost associated with plant breeding.

“Our goal was to develop a more realistic and evolutionarily informed model of the genome,” Collins elaborates. “A system that incorporates biological and evolutionary context. This refined model minimizes false positives by effectively filtering out noise and irrelevant genes.”

Reducing the Search Space for Key Genes

Consider a company researching cold-tolerant rice. A genome-wide association study initially identified 566 “genes of interest,” each requiring approximately $40,000 to investigate. This single trait could potentially incur a $20 million expense over several years, limiting accessibility to both the organizations capable of undertaking such research and the crops they choose to focus on.

“We aim to democratize this process,” Collins asserts. “Applying our technology to the same cold-tolerant rice data, we identified just 32 genes of interest. Based on our simulations and retrospective studies, we are confident that all of these genes are genuinely causal. We validated this by growing 10 gene knockouts within a three-month period.”

The Avalo model refines gene confidence levels, highlighting the most promising candidates for testing. Image Credits: Avalo

Avalo’s system initially eliminated over 90% of the genes that would have otherwise required individual investigation. The remaining genes were deemed not merely correlated with the trait, but causally linked, a finding confirmed through brief “knockout” studies where specific genes were deactivated to observe the resulting effects.

Avalo refers to its methodology as “gene discovery via informationless perturbations,” or GDIP.

Explainable AI and a Novel Learning Approach

The success of Avalo’s approach stems from the inherent capabilities of machine learning algorithms in discerning signal from noise. However, Collins emphasizes the importance of a fresh perspective, allowing the model to independently learn the underlying structures and relationships.

Crucially, the model’s results are explainable, providing justification for its conclusions rather than operating as a “black box.” This was achieved through systematic simulations where genes of interest were replaced with dummy versions, allowing the model to assess each gene’s contribution.

Avalo co-founders Mariano Alvarez (left) and Brendan Collins in a greenhouse setting. Image Credits: Avalo

Democratizing Access to Advanced Breeding Techniques

“Our technology enables us to define a minimal predictive breeding set for traits of interest,” Collins explains. “We can design the optimal genotype in silico and then conduct intensive breeding to observe its emergence.”

The reduced cost makes this approach accessible to smaller organizations, for less popular crops, or for traits with uncertain future relevance – given the unpredictable nature of climate change, it’s difficult to predict whether heat- or cold-tolerant wheat will be more advantageous in two decades.

“By lowering the financial barrier to entry, we unlock opportunities to work on climate-tolerant traits,” Alvarez adds.

Partnerships and Future Directions

Avalo is collaborating with several universities to accelerate the development of resilient and sustainable plants that might otherwise remain undeveloped. These institutions possess extensive data but often lack the resources to fully explore its potential, making them ideal partners for demonstrating Avalo’s capabilities.

These partnerships will also validate the system’s effectiveness with “fairly undomesticated” plants requiring significant improvement before large-scale implementation. For example, enhancing a naturally drought-resistant wild grain might be more efficient than attempting to introduce drought resistance into a naturally large-grained species.

Commercially, Avalo plans to initially offer its data handling services, mirroring the business model of many startups providing cost and time savings to established companies in agriculture and pharmaceuticals. Ultimately, Avalo aims to become a seed provider, bringing these improved plants to market.

Having recently emerged from the IndieBio accelerator, Avalo has secured $3 million in seed funding to expand its operations. The round was co-led by Better Ventures and Giant Ventures, with participation from At One Ventures, Climate Capital, David Rowan, and IndieBio parent SOSV.

“Brendan convinced me that launching a startup would be more rewarding than pursuing a faculty position,” Alvarez concludes. “And he was absolutely right.”