DeepMind presented results from the latest version of its AI system “AlphaFold”. According to an independent community-led assessment, the system achieves unprecedented levels of accuracy in predicting the structure of proteins. The system’s predictions were compared with experimental data to measure accuracy, reaching a score of 92.4/100 across all targets. Its average error is comparable to the width of a single atom.

“We have been stuck on this one problem – how do proteins fold up? – for nearly 50 years,” said Professor John Moult, who chairs the community which assesses protein structure prediction. “To see DeepMind produce a solution for this, having worked personally on this problem for so long and after so many stops and starts wondering if we’d get there, is a very special moment.”

Protein structure is closely linked to their functions, so the ability to predict protein structures enables a better understanding of what they do and how they work. Many grand challenges – such as developing treatments for diseases or finding enzymes which break down industrial waste – are fundamentally tied to understanding proteins. However, determining a protein structure using experimental techniques can take years of expensive work. From a database of over 200 million proteins, only a small fraction of their 3D structures has been mapped.

DeepMind presented its first iteration of AlphaFold last year, demonstrating the highest level of accuracy of all structure prediction tools assessed. Since then, the company has developed new deep learning architectures for the system.

A folded protein can be conceived as a “spatial graph”, where residues are represented by nodes and edges connect residues in close proximity. DeepMind created an attention-based neural network system – trained on 170,000 structures from the protein data bank – which attempts to interpret the structures of this graph while reasoning over the implicit graph that it is building. This graph is refined over iterations, allowing AlphaFold to generate strong predictions of the physical structure of the protein.

The system used a surprisingly modest amount of computing power: approximately equivalent to 100-200 GPUs run over several weeks.

Dr Demis Hassabis, founder and CEO of DeepMind, said: “The ultimate vision behind DeepMind has always been to build AI and then use it to help further our knowledge about the world around us by accelerating the pace of scientific discovery. For us, AlphaFold represents a first proof point for that thesis. This advance is our first major breakthrough in a long-standing grand challenge in science, which we hope will have a big real-world impact on disease understanding and drug discovery.”

DeepMind is planning to submit a paper detailing the workings of the new system to a peer-reviewed journal and is simultaneously exploring how best to provide wider access to the system in a scalable way.

The achievement has been welcomed by some of the most distinguished figures in the scientific community.

“This computational work represents a stunning advance on the protein folding problem: a 50-year-old grand challenge in biology,” said Professor Venki Ramakrishnan, Nobel Laureate and President of the Royal Society. “It has occurred decades before many people in the field would have predicted. It will be exciting to see the many ways in which it will fundamentally change biological research.”

Professor Dame Janet Thornton of the European Bioinformatics Institute commented: “What the DeepMind team has managed to achieve is fantastic and will change the future of structural biology and protein research. After decades of studying proteins, the molecules that provide the structure and functions of all living things, I awoke this morning feeling that progress has been made.”

Sundar Pichai, CEO of DeepMind’s parent company Alphabet (and its subsidiary Google), added: “This is an incredible AI-powered breakthrough in protein folding, which will help us better understand one of life’s most fundamental building blocks. This huge leap forward from DeepMind has immediate practical implications, enabling researchers to tackle new and difficult problems, from future pandemic response to environmental sustainability.”

DeepMind presented results from the latest version of its AI system “AlphaFold”. According to an independent community-led assessment, the system achieves unprecedented levels of accuracy in predicting the structure of proteins. The system’s predictions were compared with experimental data to measure accuracy, reaching a score of 92.4/100 across all targets. Its average error is comparable to the width of a single atom.

“We have been stuck on this one problem – how do proteins fold up? – for nearly 50 years,” said Professor John Moult, who chairs the community which assesses protein structure prediction. “To see DeepMind produce a solution for this, having worked personally on this problem for so long and after so many stops and starts wondering if we’d get there, is a very special moment.”

Protein structure is closely linked to their functions, so the ability to predict protein structures enables a better understanding of what they do and how they work. Many grand challenges – such as developing treatments for diseases or finding enzymes which break down industrial waste – are fundamentally tied to understanding proteins. However, determining a protein structure using experimental techniques can take years of expensive work. From a database of over 200 million proteins, only a small fraction of their 3D structures has been mapped.

DeepMind presented its first iteration of AlphaFold last year, demonstrating the highest level of accuracy of all structure prediction tools assessed. Since then, the company has developed new deep learning architectures for the system.

A folded protein can be conceived as a “spatial graph”, where residues are represented by nodes and edges connect residues in close proximity. DeepMind created an attention-based neural network system – trained on 170,000 structures from the protein data bank – which attempts to interpret the structures of this graph while reasoning over the implicit graph that it is building. This graph is refined over iterations, allowing AlphaFold to generate strong predictions of the physical structure of the protein.

The system used a surprisingly modest amount of computing power: approximately equivalent to 100-200 GPUs run over several weeks.

Dr Demis Hassabis, founder and CEO of DeepMind, said: “The ultimate vision behind DeepMind has always been to build AI and then use it to help further our knowledge about the world around us by accelerating the pace of scientific discovery. For us, AlphaFold represents a first proof point for that thesis. This advance is our first major breakthrough in a long-standing grand challenge in science, which we hope will have a big real-world impact on disease understanding and drug discovery.”

DeepMind is planning to submit a paper detailing the workings of the new system to a peer-reviewed journal and is simultaneously exploring how best to provide wider access to the system in a scalable way.

The achievement has been welcomed by some of the most distinguished figures in the scientific community.

“This computational work represents a stunning advance on the protein folding problem: a 50-year-old grand challenge in biology,” said Professor Venki Ramakrishnan, Nobel Laureate and President of the Royal Society. “It has occurred decades before many people in the field would have predicted. It will be exciting to see the many ways in which it will fundamentally change biological research.”

Professor Dame Janet Thornton of the European Bioinformatics Institute commented: “What the DeepMind team has managed to achieve is fantastic and will change the future of structural biology and protein research. After decades of studying proteins, the molecules that provide the structure and functions of all living things, I awoke this morning feeling that progress has been made.”

Sundar Pichai, CEO of DeepMind’s parent company Alphabet (and its subsidiary Google), added: “This is an incredible AI-powered breakthrough in protein folding, which will help us better understand one of life’s most fundamental building blocks. This huge leap forward from DeepMind has immediate practical implications, enabling researchers to tackle new and difficult problems, from future pandemic response to environmental sustainability.”