DeepMind’s AlphaFold Predicts Over 200 Million Protein Shapes in AI Breakthrough

Only a year after publishing a database of 350,000 protein structures, DeepMind’s Alphafold AI platform announced an update that will include 200 million predicted protein structures, all openly available to the public. The breakthrough is a testament to AI’s strength and potential to assist and advance biological understanding and scientific development. 

Broadening the Scope of the Project to Include Every Known Protein

The initial release of AlphaFold included several hundred thousand predicted protein structures and grew to over a million over a year. The researchers at DeepMind, Google’s London-based AI company, focused primarily on known human proteins and proteins that would be useful for research, like those found in mice and other research-oriented organisms. 

With the deep learning algorithm implemented in AlphaFold’s platform, the company is expanding the database to include protein structures from millions of animals, plants, fungi, and other organisms. The additional protein structure predictions will consist of nearly all proteins known to science. 

Before AI advancements, scientists could spend months or even years trying to manually construct a protein’s structure through costly methods like X-ray crystallography. Now, AlphaFold can predict a protein’s structure in a matter of seconds and determine the accuracy of the prediction. 

“AlphaFold is the singular and momentous advance in life science that demonstrates the power of AI. Determining the 3D structure of a protein used to take many months or years, it now takes seconds,” Founder and Director of the Scripps Research Translational Institute, Eric Topol, said, “AlphaFold has already accelerated and enabled massive discoveries, including cracking the structure of the nuclear pore complex.”

Related Article: Ushering In A New Generation Of AI-Focused Investments Opportunities

Harnessing the Power of Revolutionary Computational Technology

The applications of a huge protein structure database are nearly limitless, given that proteins are truly the building blocks of life, inherent in processes in every living organism on the planet. Predicting protein structures could help in understanding large-scale problems like the evolution of proteins to more focused areas like individual drug discovery for unsolved diseases. 

A honeybee researcher from the Norwegian University of Life Science, Vilde Leipart, used AlphaFold’s technology to understand honeybee reproduction and immunity better. Leipart used AlphaFold to predict the structure of vitellogenin, a protein that supports reproduction. He said that even though scientists have had a general understanding of the protein since the 1990s, having a 3D model of the protein helped him understand key processes of the protein that would have taken years to develop without machine learning technology. 

Protein structure prediction is also of particular interest in the drug discovery realm. By understanding how proteins are structured, researchers can now look at how certain proteins interact with those around them. In doing so, some researchers, like Karen Akinsanya at Schrödinger in New York, can use other software programs to simulate how proteins interact with each other and potentially develop drugs that only target specific proteins, leaving those around them untouched. 

While AlphaFold’s technology is not the answer to all of biology’s woes, it might be a piece to a much larger puzzle for scientists worldwide to use and collaborate with to find solutions to some of humanity’s biggest problems. DeepMind CEO Demis Hassabis warned that there is still a lot of work to be done in biology and chemistry to better leverage AlphaFold’s platform but said encouragingly, “We’re at the beginning of a new era of digital biology.” 

Related Article:

Ultima Genomics Promises $100 Genome After Raising $600 Million

Top iPS Cell Technologies at Affordable Prices – Keynote by Nobel Laureate Shinya Yamanaka at BIO Asia-Taiwan

Author

Reed Slater