Artificial intelligence is increasingly being deployed across the food system from farm to fork, optimising everything from seed selection to reformulation and R&D.
But Google DeepMind is coming at the problem from a different direction.
Understand and map biology’s building blocks, then see where that leads.
“If we can solve the very fundamental problems in a given scientific space, we can then unlock all of these downstream use cases in a much more optimal, systematic fashion,” says Director Ramine Tinati, speaking at the Asia-Pacific Agri-Food Innovation Summit 2026 in Singapore.
The scientific sandbox DeepMind has built spans three areas: protein structures through AlphaFold; understanding how changes in DNA can affect biological function through AlphaGenome; and a longer-term ambition to build virtual cells that could allow more biological experimentation to take place using AI and computation.
All three could have implications for the food system. Here’s how.
1. Building crops that can fight disease
To understand AlphaFold’s impact, you need a quick biology lesson.
Proteins are chains of amino acids that fold into complex three-dimensional structures. That shape is fundamental to what a protein does, so being able to predict it can help scientists understand biological processes and identify potential targets for new treatments and technologies.
This is where AlphaFold changed the game. Rather than scientists spending months or even years determining individual protein structures experimentally, DeepMind’s system can predict them computationally. Its database now contains predictions for hundreds of millions of protein structures.
And the applications go beyond human health.
“We’re using it to create more disease-resistant crops,” says Tinati, referring to work being conducted by researchers at the Sainsbury Laboratory in the UK, which could help develop plants with stronger immunity.
DeepMind is also applying its genomics capabilities to crop resilience.
Tinati points to work involving sorghum, a major food and animal-feed crop, where he says some of DeepMind’s AlphaGenome technologies are being used to map varieties that may be more resilient to external conditions.
The resulting resource is being made openly available to researchers.
“We’re making that available for public good,” says Tinati. “It’s an open atlas available for researchers to consume and hopefully optimise the plants or seeds that they are growing.”
2. Helping farmers produce more in a less predictable climate
The second application moves out of the laboratory and into the field.
DeepMind has been working with smallholder farmers and soil scientists in India on how limited farm-level data can be combined with wider environmental information to make better decisions about what to grow and when.
The work brings together information about individual land parcels with longitudinal data and factors including soil conditions, weather and a changing climate.
The ambition is decidedly practical.
“As a smallholder farmer, I’ll then know what seeds to grow, when is optimal to harvest and when to rest my plots to support maximum yield,” says Tinati.
Having begun in India, Tinati says the programme is now being scaled to 11 countries.
And the challenge becomes more pressing as growing conditions become less predictable.
“We are in a world where climate is changing and we are seeing more and more extreme deltas in the weather,” says Tinati. “That also means that the foods or crops we have today need to be more resilient in the future to be able to continue to deliver the same yields, if not more.”
3. Making experimentation itself faster
Perhaps the biggest impact, however, could be less visible.
DeepMind isn’t only developing models that answer particular scientific questions. It is also looking at whether AI can accelerate the process of science itself.
That means shortening the journey from idea to hypothesis, experiment, result and action.
“A lot of the time, whether it’s data processing, whether it’s doing the computational process or being able to analyse results and accelerate the insights from a hypothesis to result, that takes a lot of time,” says Tinati.
One of the tools being developed to tackle this is DeepMind’s AI co-scientist.
Researchers can begin with an idea and use the system to develop experimental approaches, exploring thousands of potential permutations and drawing on scientific literature, including ideas from across different disciplines.
The objective is not simply to automate existing scientific processes, but to dramatically widen the number of possibilities scientists can investigate before committing resources to physical experiments.
And that feeds into a much bigger ambition at DeepMind: the virtual cell.
The long-term vision is for researchers to be able to conduct far more biological experimentation computationally, reducing the amount of work that needs to move immediately into a wet lab.
“Being able to perform a variety of different biology-related experiments and life sciences experiments in computation means you don’t have to go to the wet lab, or the process from computation to wet lab is shortened,” says Tinati.
AI can also spot possibilities human scientists may miss.
“Scientists have a certain judgement and biases in the way we do experimentations, but the tools allow us to systematically prioritise and rank things that we may have missed without these tools to augment us,” says Tinati.
That doesn’t mean humans are destined to be banned from the lab, however.
“It’s about augmentation. It’s not about replacing the scientist.”
Discover more from PMN S.P.O.R.T.S - A PRIME MEDIA NETWORK BRAND
Subscribe to get the latest posts sent to your email.
