Next Gen AI: How Protein Design Could Transform the World

Anna Tamara
August 1, 2023
08 min
AI, Protein Design, Protein Folding.
AI, Science and BioTech.

Away from the language-model buzz, the latest breakthroughs for artificial intelligence are in generative biology. 

AI solving the protein folding problem is set to transform science, industry and society. Protein design has long been a grand challenge in biology – it takes a PhD student their entire PhD to fold one protein. Scientists have dedicated their entire lives to better understanding these vastly complex structures. 

When DeepMind unveiled their AlphaFold AI system in November 2020, scientists declared this would change everything. Now AI can predict the 3D structure of proteins based on amino acid sequences. Some of the models in development can generate millions of proteins in a few days – essentially ‘unlocking’ biology. Today’s discoveries, including those facilitated by DeepMind’s AlphaFold2 and OmegaFold, are propelling the field further. Researchers are uncovering proteins not produced in nature with ever higher success rates. 

We’re at the beginning of understanding how this AI will transform science. 

It’s a landmark shift in our current moment. With potential to accelerate scientific discovery in any field. Reimagining medicine, engineering and climate solutions – especially nuclear fusion, which some say could solve the climate crisis. As we anxiously anticipate the evolution of AI and its exponential impact on society, how much is there to hope and fret over with this latest news? What could the future look like if its application is as powerful as predicted?

It could take us beyond what’s possible in the natural world.

The machine-learning architecture builds on AI systems like DALL-E 2, known for generating realistic images from natural language prompts. Broadly, these models are trained to spot patterns in proteins and predict new sequences to solve a problem. Interestingly, researchers have been impressed by the AI’s humility when it ‘doesn’t know’ the answer. If faced with a task it can’t solve, instead of producing improbable proteins, it generates the closest possible solution. Along with recent successes in test tube experiments, experts have growing confidence that the real world applications of these proteins are viable.

The impact could be monumental. Starting in medicine, where shortcuts for scientists help speed up drug development. 

Skipping phases of laborious, time-consuming experiments, it could take years off timelines for drug discovery. From identifying the structure of proteins effective in treating multiple diseases, to the tricky part – adapting these into usable drug treatments at scale. As scientists work to break through this limitation, US researchers are testing tailor-made, functional proteins that they could produce in live cells. 

If successful, cutting costs this way could bring big pharma’s benefits to more of humanity, with particular interest in treatments for neglected diseases, like parasitic infections in underdeveloped countries. Further ahead in the long term vision, DeepMind’s RFdiffusion model hopes to uncover proteins effective in tackling cancer, influenza, and even the next pandemic. The team is working with biologists on pathogens that could cause a future pandemic, with the eventual aim to facilitate vaccines. Will we find a way out of a pandemic before it begins?

We could soon be stretching the limits of manufacturing, with new materials reshaping design and engineering in unimaginable ways.

MIT is driving research to design proteins with specific structural objectives. Similar to those existing in nature but with new properties, such as stiffness and elasticity. The possibilities are so vast, its scope in these early phases is unclear. But it could meet our most pressing design needs. Replacing materials made from petroleum or ceramics, for example, and producing them with a much smaller carbon footprint. Might this be a driver in weaning us off the earth’s natural resources? Food scarcity, waste and malnutrition could be reduced too, with biologically-inspired food coatings to keep produce fresh for longer. Further enhanced by emerging innovations in biotechnologies, we could be headed for lab-synthesized solutions to the biggest problems facing us today.

As well as designing sustainable new biomaterials, these proteins could one day capture solar energy, remove carbon from the atmosphere, or find new clean energy sources. 

In the fight to save us from pollution, scientists are looking into enzymes that ‘eat’ plastic like a food source. Breaking down plastic into its building blocks can allow us to recycle it endlessly and accelerate environmental clean up. We could find proteins that ‘eat’ carbon out of the atmosphere, or even faster routes to viable nuclear fusion reactors. Touted as the ‘holy grail’ clean energy of the future, nuclear fusion still looks to be decades away. Unlocking its power sooner could tip the scales in our efforts to halt climate change. 

Of course, AI’s promise also comes with peril.

As pioneers like DeepMind CEO Demis Hassabis remind us, these powerful new tools carry unforeseen risk. In the wrong hands, this technology could be used to game the financial market, for example, or even design sophisticated biological warfare. With its capacity to generate deadly pathogens, it can uncover around 40,000 possible chemical weapons in 6 hours, according to a recent paper

This is without speculating on the AI doom scenario, where artificial general intelligence (AGI) surpasses human intelligence and turns against us in unexpected ways. However uncertain our AI future, Hassabis is also hopeful AGI could drive us toward ‘radical abundance’: a world with more equality, more resources, and more wealth. If so, these miracle proteins might just give us a head start.