December 3, 2020

'This will change medicine': How DeepMind is answering one of biology's biggest questions

Daily Briefing

    On Monday, Google's DeepMind unit released data showing that its artificial intelligence (AI) program AlphaFold can accurately and quickly predict the 3D structure of a protein—a development that addresses one of the longest-enduring challenges in the field of biology and could help researchers create more effective drugs.

    Related: Artificial Intelligence Resource Library

    AlphaFold shows promising results

    As part of a biennial protein-structure prediction challenge called Critical Assessment of Structure Prediction, or CASP, DeepMind set out to train AlphaFold to predict the 3D shape a protein would take based on the protein's amino-acid sequence. To do so, DeepMind provided AlphaFold with about 170,000 proteins acquired from a public repository of protein structures and sequences.

    AlphaFold compared those sequences, looking for amino-acid pairs that typically ended up close together in folded protein structures. AlphaFold then utilized that data to estimate how far apart pairs of amino acids would be in structures not yet known.

    For the CASP competition, AlphaFold was challenged with making predictions for proteins whose structures were already known, but not yet publicly disclosed. Scientists compared AlphaFold's predictions with the known structure of the proteins to assess the technology's effectiveness.

    DeepMind on Monday announced that AlphaFold scored around a 90 out of 100 on predicting protein structures that were determined to be moderately difficult, while its competitors scored around 75 out of 100. According to the MIT Technology Review, a score higher than 90 "means that any differences between the predicted structure and the actual structure could be down to experimental errors in the lab rather than a fault in the software," or that the software's prediction "is a valid alternative configuration to the one identified in the lab, within the range of natural variation."

    Almost two-thirds of AlphaFold's predictions were comparable to the quality of structure determinations generated by other means, such as X-ray crystallography.

    'This will change medicine'

    Proteins are the building blocks of life, and how they work is determined by their 3D shape, Nature reports. According to the MIT Technology Review, being able to determine what a protein does "is key to understanding the basic mechanisms of life."

    Generally, experiments conducted in laboratories—most often involving X-ray crystallography—have been the primary mechanism used for determining a protein's structure. But those experiments are complex, and scientists have pondered whether strings of amino acids also could be used to predict a protein's eventual structure. Although researchers over the years have attempted to create methods for predicting a protein's structure using amino acids, none of those methods of have proved accurate—until now.

    John Moult, a computational biologist at the University of Maryland and co-founder of CASP, said AlphaFold's ability to accurately predict proteins' structures effectively answers one of biology's longest-enduring challenges. "This is a big deal," he said. "In some sense the problem is solved."

    Mohammed AlQuraishi, a computational biologist at Columbia University, said AlphaFold "will be very disruptive to the protein-structure-prediction field."

    "I suspect many will leave the field as the core problem has arguably been solved. It's a breakthrough of the first order, certainly one of the most significant scientific results of my lifetime," AlQuraishi added.

    Andrei Lupas, a biologist at the Max Planck Institute for Developmental Biology, said AlphaFold was able to predict the structure of a protein his lab hadn't been able to figure out for more than a decade—and the technology could have significant effects for health care.

    "This will change medicine," Lupas said. "It will change research. It will change bioengineering. It will change everything."

    Experts say AlphaFold could accelerate drug discovery. According to The Economist, the AI also could be used to determine whether existing drugs are able to treat conditions that scientists aren't aware they're able to treat. AlphaFold also could lead to improvements in synthetic biology, by allowing researchers to more quickly develop human-designed proteins that catalyze chemical reactions that help bodies function.

    Further, AlphaFold could have an impact on the novel coronavirus pandemic, as it was able to predict a number of the proteins found in the virus that weren't previous known, Becker's Health IT reports.

    "We're definitely going to want to spend some time kicking the tires," Ewan Birney, deputy director of the European Molecular Biology Laboratory, said. "But when I first saw these results, I nearly fell off my chair."

    Demis Hassabis, co-founder and CEO of DeepMind, said the company intends to make AlphaFold usable by other scientists, and Hassabis believes the technology could be used for protein design and drug development.

    "We're just starting to understand what biologists want," Hassabis said (Ross, STAT News, 11/30; Callaway, Nature, 11/30; Heaven, MIT Technology Review, 11/30; The Economist, 11/30; Dyrda, Becker's Health IT, 12/1).

    Learn more: The path to artificial intelligence


    A new wave of AI-powered capabilities is likely to improve—and even—transform health care operations, but success with these new technologies requires a strong foundational analytics program.

    Health care organizations must ensure they have the necessary data sources, architecture, governance, talent, leadership, and data-driven culture for their programs to deliver consistent value. With all of the necessary assets in place, health care organizations will then be ready to put analytics and AI into practice.

    Download Now
    Cookies help us improve your website experience. By using our website, you agree to our use of cookies.