Generative-AI: Dreaming up proteins

Back in 2021, artificial intelligence solved a mystery that had been slowing the progress of science for almost a century: how to figure out a protein's structure from its amino acid sequence.

The scientific roadblock, called the "protein folding problem," was solved by AlphaFold, an AI tool from Google's DeepMind laboratory.

Proteins are the building blocks of life — literally, everything that happens in life or nature as a whole depends on proteins. They include antibodies that fight illnesses, hemoglobin that carries oxygen in red blood cells, and enzymes.

A protein's structure is crucial for its function. It's the structure of hemoglobin that helps it carry oxygen in red blood cells, and the structure of enzymes, like amylase, that helps them chew up starches to make sugars.

"You're working in the dark if you don't know what the protein looks like. If you want to modify a protein, like with a drug, it's much clearer when you have the structure," said Kathryn Tunyasuvunakool, a research scientist at DeepMind who was part of the team that created AlphaFold.

AlphaFold accurately predicts the folded shape of proteins from amino acid sequences. You can feed it any sequence of amino acids and it pops out the protein shape instantly.

Tunyasuvunakool said AlphaFold has made a huge impact on science, accelerating research in every field of biology, as well as other technology sectors.

It has been used to develop malaria vaccines and drugs to treat cancer, develop plastic-digesting enzymes and design proteins that can convert sunlight into fuel with high efficiency.

AlphaFold's usage speaks volumes for its popularity among scientists. Two years on, the AlphaFold Protein Structure Database has been used by more than 1.2 million researchers around the world. 

AI tools create entirely new proteins

AlphaFold was just the beginning of our using AI with proteins. Since then, scientists have got even more creative.

Mohammed AlQuraishi, a molecular biologist and AI expert at Columbia University in the US, has taken the idea behind AlphaFold to the next level — if you can solve the problem of protein folding, why not create entirely new proteins?

AlQuraishi came up with Genie, a generative AI model of protein design that uses digital art techniques to create custom proteins. The result is a tool that can dream up entirely new proteins that have never existed before in nature.

Genie was released in July 2023 in a pre-print study (not yet peer-reviewed) and is one of two similar AI tools presented in 2023 by other research groups.

Genie was repurposed from AlphaFold, essentially merging its capabilities with generative art image programs, like MidJourney.

AlQuraishi and his team trained Genie with data about the charges and structures of amino acids, and how they interact to form proteins.

"It's like if you're trying to make pictures of people [with AI programs]. You give it lots and lots of examples of faces. Initially it picks out high level features like shape, but then learns finer features like hair and 'faceness.' After learning the features of faceness, it can eventually generate new faces," said AlQuraishi.

It's the same concept with protein design, continued AlQuraishi — Genie first learns simple features of proteins, then generates precise atomic placements to form new proteins.

But like AI-generated faces, Genie creates proteins that have never existed in nature. They are completely made up.

Digital proteins may transform science

Tunyasuvunakool said generative AI tools like Genie could be a major boon for scientific progress.

AI generative protein design tools help by linking the role of a protein's structure with its function. For example, it can help scientists understand how the buildup of Tau plaques in neurons (protein fragments in the brain) contribute to Alzheimer's disease. 

"Another example is understanding basic questions of evolution. AI can shed light on how protein structures evolved over 4 billion years," said AlQuraishi.

The second benefit is in medical science. Say you want to design a molecule to treat a disease, such as designing a molecule that breaks down those Tau plaques to cure Alzheimer's.

"When you're designing a drug, it's very helpful to know the structure of the protein you're targeting. It tells us how the drug binds to protein targets and how it behaves during the interaction," said AlQuraishi.

AlQuraishi described himself as feeling "very bullish" that AI tools will transform medical science, but also said he thinks that more immediate benefits will be seen in energy, the environment, or agricultural sectors.

"We could design new enzymes that help break down pollutants or plastics. It's here I think we'll see a quicker impact rather than medical science because the regulation is lighter," said AlQuraishi.

AI tools can't model mobile proteins, yet

There's one fundamental problem with AI protein structure tools, including  AlphaFold and Genie: they only design proteins that are static and rigid.

Real proteins — those found in nature — don't work that way. Real proteins can change shape and adapt to different contexts.

Take, for example, proteins on the surface of neurons, called ion channels, open and close like gates to let sodium and potassium in or out of the cell. 

"We can't design that right now, it's something that remains beyond the realm of existing AI tools, including ours. This is the next frontier and what we're working to solve next," said AlQuraishi.

But the joy of AI is you can go back and train it to learn new things. Real world data can be fed back into the AI, such as how a new protein-based drug is metabolized and functions in the body, or how efficient a plastic-eating enzyme works in a recycling plant. The AI can relearn how protein structures impact their function in the real world.

Edited by: Zulfikar Abbany.