This image shows part of a protein from (Cellulophaga baltica crAss-like) phage phi14:2, a virus that infects bacteria. This phage is part of a group of viruses that includes the most abundant virus in the human gut. The protein itself (gp66) is a newly-discovered DNA-dependent RNA polymerase.

AI network unlocks secrets of protein structures

An algorithm has provided the key to predicting the three-dimensional shapes of folded proteins, opening new pathways for science, including in the treatment of neurodegenerative diseases and the discovery of new drugs. 

Molecular biologists announced a major breakthrough in the mapping of proteins in the human body on 30 November 2020, after an artificial intelligence (AI) network managed to deduce the three-dimensional structure of a protein from the sequence of its amino acids. The AlphaFold 2 “deep learning” program, developed by Google subsidiary DeepMind, demonstrated the ability to determine the shape of protein folds as part of an international research competition, the 14th Community Wide Experiment on the Critical Assessment of Techniques for Protein Structure Prediction (CASP14). These researchers have been trying systematically since 1994 to find ways of charting protein structures by other means than X-ray crystallography and advanced electron microscopes. Contrary to the famous industrial design maxim that “form follows function”, at the molecular level, it is the physical configuration of a protein’s amino acids that determines its function. Conversely, errors in protein folding can lead to neurodegenerative diseases. Scientists have accordingly been trying for more than half a century to find ways of predicting the folded structures of proteins with accuracy.  

Unprecedented accuracy 

The DeepMind team used AI to predict the distances between individual amino acid pairs based on physical and genetic parameters, and then to predict the entire three-dimensional structure of the protein. Within the CASP14 competition, each research team was given proteins to work on whose structure, though unknown to them, had already been established by experimental means using nuclear magnetic resonance imaging. AlphaFold 2 predicted the folding patterns with an accuracy of about 90 out of 100 points, compared to other teams’ scores of around 75.

Following the announcement, several researchers quoted by media outlets expressed astonishment at the breakthrough, noting that the discovery, while long anticipated, had come much faster than expected. The ability to understand the precise folding structures will likely provide a boost to drug discovery and protein design, opening up many new avenues for treatments and therapies – decades sooner than many scientists had dared to hope. But the success of the “deep learning” approach embodied by the AlphaFold algorithm also confirms that AI has brought us to the cusp of a new era in scientific research, with exciting applications far beyond the field of molecular biology. 

Are you an expert in your field?

Supertrends is a technology and media company dedicated to uncovering the future. At Supertrends, we believe in expertise. In order to maintain the quality and value of our articles, reports and timeline, we go straight to the sources of knowledge about the future – experts. Supertrends experts are highly qualified individuals who are knowledgeable about, and actively working to create, the innovations that will shape our future. If this sounds like you, you can read about the application process as well as the benefits of being a Supertrends expert here.

Thank you to Feedspot for listing us in the top 50 futurism blogs and websites to follow in 2020.

© 2020 Supertrends


Senior, A.W. et al. (2020) “Improved protein structure prediction using potentials from deep learning”. Nature 577:706-10.

Calloway, E. “‘It will change everything’: DeepMind’s AI makes gigantic leap in solving protein structures”,, 30 November 2020, available at:

AlphaFold, Artificial Intelligence, Deep Learning, DeepMind, Molecular Biology, Neurodegenerative Diseases, Protein Folding, Protein Structures

Chris Findlay

I'm a journalist, editor, and translator based in Zurich, Switzerland. I write about technology and future timelines at, where I also help expand the community as Expert Relationship Manager.

Sign up for our Newsletter


For questions or more information

Contact us


Supertrends AG
Østergade 40, 2.
1100 Kbh. K 


Supertrends AG           
Erlenstrasse 16
6300 Zug

Join Us