This past week Alphabet, the parent company of Google, announced that its DeepMind group has solved a long-standing grand challenge in the scientific community around protein structure prediction.
In 1972, scientist Christian Anfinsen predicted that a protein’s structure could be determined by its amino acid sequence. However, figuring out that sequence is immensely difficult since there are near-infinite ways in which a protein can fold. This led Cyrus Levinthal to postulate that calculating all the known configurations would take longer than the age of the known universe if our only way there was brute force calculation (a problem often referred to as Levinthal’s Paradox).
Thankfully, now we won’t have to wait forever (literally) since DeepMind’s DeepFold AI can predict protein structure to the width of an atom within days. My former Gigaom colleague, Darrell Etherington, explained the feat in a post on Techcrunch:
The test that AlphaFold passed essentially shows that the AI can correctly figure out, to a very high degree of accuracy (accurate to within the width of an atom, in fact), the structure of proteins in just days — a very complex task that is crucial to figuring out how diseases can be best treated, as well as solving other big problems like working out how best to break down ecologically dangerous material like toxic waste.
As many in the world of future food development know, AI is becoming an increasingly important tool to accelerate the development of new proteins and other food-building blocks. Companies like Climax Foods are embracing machine learning to help them develop new approaches to making food like plant-based cheese. And now, with DeepMind’s advances in protein folding prediction paths, we can expect AI to become an even more important tool in new food development.
To get a feel for what the impact of this milestone might be, I asked Sudeep Agarwala, a synthetic microbiologist for Gingko, what he thought DeepMind’s work could possibly mean for food:
“There’s so much that can happen with this,” Agarwala told me. “Think about different textures or mouth feels with food proteins that the AI can design. Or even different amino acid contents for the proteins. And that’s just for the end proteins (if you want the protein as the end product).”
“If we’re engineering proteins inside the cell to produce a small molecule (a flavor, a fragrance, a fat) we can think about making completely new enzymes with multiple functionalities. So something like a 5-step process might be condensed. This has the potential to simplify so much of the metabolic engineering we do to produce these products.”
Agarwala also believes that using technology like that of DeepMind’s will also be way more effecient in terms of resources and have a much smaller ecological footprint:
“Even if the proteins aren’t ultimately going to be consumed, being able to rationally design enzymes that are involved in producing small molecules (think fats, fragrances, or flavors) is really exciting to think about,” said Agarwala. “This isn’t small potatoes: think about how much energy and resources goes into making vanilla or saffron, for example. Being able to do this more efficiently will provide a less ecologically expensive way to produce these materials.”
Just as disease and virology experts are buzzing about what this new milestone could mean for their work, it’s clear food scientists like Agarwala are equally as excited.
“It’s such an exciting time to be a biologist!” said Agarwala.
Yes, it is.
