Deepmind AI AlphaFold 50 Year Old Grand Challenge of Protein Structure Prediction

Two examples of protein targets in the free modeling category. Alphafold predicts highly accurate structures measured against experimental results. Sincerely: Deepamind

DeepMind develops AI solutions to the 50-year-old protein challenge, creating the potential to accelerate biological research.

In a major scientific advance, the latest version of DeepMind’s AI system AlphaFold has been recognized as a solution to the 50-year grand challenge of predicting protein structure, often referred to as the ‘protein folding problem’ according to an Independent is. Evaluation. This breakthrough may accelerate biological research in the long run, opening up new possibilities in other fields of disease understanding and drug discovery.

Today, CASP14 results show that DeepMind’s latest alphafold system achieves unique levels Accuracy In structure prediction. The system is able to determine highly-precise structures in just a few days. CASP, the Critical Assessment of Protein Structure Prediction, is a biennial community-driven assessment that began in 1994, and the gold standard for estimating predictive techniques. Participants should visually estimate the structure of the protein that is only recently – or in some cases not yet – determined experimentally, and wait for their predictions to be compared to experimental data.

CASP uses the “Global Distance Test (GDT)” metric to assess accuracy, ranging from 0–100. The new AlphaFold system achieves an average score of 92.4 GDT overall across all targets. The average error of the system is about 1.6 Angstrom – about the width of one Atom. According to CASP co-founder and president Professor John Mool, a score of around 90 GDT is informally considered to be competitive with results obtained by experimental methods.

Professor John Mool, co-founder and president of CASP University, Maryland, said:

“We are stuck on this one problem – how proteins fold for over 50 years. Deepmind personally worked on this problem for so long to see the creation of a solution for it, and after stopping so much and started wondering if we would ever get there, this is a very special moment. “

Why protein structure prediction matters

Proteins are essential for life and their size is closely associated with their functions. The ability to predict protein structures provides a better understanding of what they do and how they work. The main database currently contains more than 200 million proteins and only part of their 3D structures have been mapped.

A major challenge is the astronomical number in which the protein can bend theoretically before settling into its final 3D structure. There are many major challenges facing society, such as developing treatments for diseases or finding enzymes that break down industrial waste, is fundamentally tied to proteins and the role they play. Determination of protein shapes and functions is a major area of ​​scientific research, primarily using experimental techniques that according to the structure can take years of laborious and laborious work, and require the use of multi-million dollar specialized equipment is.

DeepMind’s Approach to the Protein Folding Problem

This success builds on DeepMind’s first entry into CASP13 in 2018, where the early version of Alfold achieved the highest level of accuracy among all participants. Now, DeepMind has developed new deep learning architectures for CASP14, drawing inspiration from the fields of biology, physics and machine learning, as well as the work of many scientists in the protein folding field over the last half century.

A folded protein can be thought of as a “spatial graph”, where the residues are nodes and the edges are closely joined to the residues. This graph is important for understanding the physical interactions within proteins, as well as their evolutionary history. For the latest version of AlphaFold used in CASP14, DeepMind created an attention-based neural network system, trained end-to-end, that attempts to explain the structure of this graph, while the underlying Arguments on the graph that it is constructing. It uses sequentially related sequences, multiple sequence alignments (MSAs), and amino representations. Acid Add the residuals to refine this graph.

By iterating this process, the system develops strong predictions of the underlying physical structure of the protein. Additionally, alphafold can predict which parts of each predicted protein structure are reliable using internal confidence measurements.

The system was trained on publicly available data consisting of ~ 170,000 protein structures from a protein data bank, using relatively modest amounts compared to modern machine learning standards – about 128 TPUv3-cores (about ~ 100- Equivalent to 200 GPU) a few weeks.

Potential for real world impact

DeepMind is excited to collaborate with others to learn more about AlphaFold’s potential, and the AlphaFold team is looking at how protein structure prediction can contribute to understanding certain diseases with certain expert groups.

As one of the many tools developed by the scientific community, there are indications that protein structure prediction may be useful in future epidemic response efforts. Earlier this year, DeepMind predicted several protein structures SARS-CoV-2 Viruses, and experimentally accelerated experimental work, have now confirmed that AlphaFold achieved high accuracy on its predictions.

AlphaFold is one of the most important advances to date. But with all scientific research, there is still much work to be done, including finding out how many proteins are complex, how they interact DNA, Royal army, Or small molecules, and how to determine the exact location of all amino acid side chains.

Like its earlier CASP13 alphafold system, DeepMind plans to submit a course review journal in due time to expand the functioning of the system, as well as discover that the system is scalable How to provide broad access to

Alpha Fold breaks new ground in demonstrating astonishing potential for AI to aid fundamental scientific discovery. Deepind looks forward to collaborating with others to unlock that potential.

Statement from independent scientists:

Professor Venky Ramakrishnan, Nobel Laureate and President of the Royal Society
“This computational work represents an astonishing advance on the 50-year grand challenge in biology on the protein-folding problem. This has happened decades ago when many people in the area would have predicted it. It will be exciting to see many ways to fundamentally change biological research. “

Professor Dame Janet Thornton, Director Emeritus and Senior Scientist, EMBL-EBI
“What Deepmind’s team has achieved is fantastic and will transform structural biology and protein research in the future. After studying proteins for decades, the molecules that provide the structure and functions of all living things, I realize this morning that progress has been made. “

Arthur D. Levinson, PhD, founder and CEO Calico, former president and CEO, Genentech
“AlphaFold is a generation in advance, predicting protein structures with incredible speed and precision. This leap further demonstrates how computational methods are designed to transform research in biology and promise much more to speed up the process of medicine. “

Professor Andrei Lupus, Director, Max Planck Institute for Developmental Biology
“The amazingly accurate model of alphafold has allowed us to resolve a protein structure that we had been locked into for a decade, canceling our attempt to understand how signals are transmitted across the cell membrane . “

Professor Ivan Birney, Deputy Director General EMBL, Director EMBL-EBI
“When I saw these results, I fell from my chair. I know how rigorous CASP is – it basically ensures that computational modeling must perform on the daunting task of folding ab-initio proteins. It was shocking to see that these models could perform so accurately. There will be many aspects to understand, but this is a big advancement for science. “

Statement from DeepMind / Alphabet:

Demis Haasbees, PhD, Founder and CEO, DeepMind
“The ultimate approach behind DeepMind has always been to build AI and then use it to help advance our knowledge of the world around us by accelerating the pace of scientific discovery. For us AlphaFold represents a first proof point for that thesis. This is a long-standing major challenge in advance science, which we hope will have a major real impact on disease understanding and drug discovery. “

Pushmeet Kohli, PhD, Head of AI for Science, Deepmind
“These incredible results are testament to DeepMind’s unique research philosophy – bringing mission-focused, multi-disciplinary teams together to target ambitious scientific goals. Critical evaluations such as CASP are critical to furthering research progress, and we look forward to building on this work, deepening our understanding of proteins and biological mechanisms, and opening up new avenues of exploration. “

John Jumper, PhD, AlphaFold Lead, DeepMind
“Protein biology is hypothetically complex and defines simple characterization. Our team’s work demonstrates that machine learning techniques are finally able to meet the complexity of describing these incredible protein machines, and we’re really excited to see what new breakthroughs in both human health and fundamental biology will bring. “

Katherine Tunisuvanakul, PhD, Science Engineer, DeepMind
“The ability to predict high-precision protein structures with AI can change how we approach biology, with potential applications in drug design and bioremediation. Especially for experimentally challenged proteins, good predictive technology can bring huge changes. “

Sundar Pichai, CEO, Google and Alphabet
“This is an incredible AI-driven breakthrough in protein folding, which will help us better understand one of life’s most basic building blocks. This huge leap beyond DeepMind has immediate practical implications, enabling us to deal with new and difficult problems in future from epidemic response to environmental sustainability. “

For more information on this topic read DeepMind AI Solution in the 50-Year-Old Science Challenge “Can revolutionize medical research”.

About Deepamind

DeepMind is a multi-disciplinary team of scientists, engineers, machine learning experts and more, working together to research and build secure AI systems that learn to solve problems and learn scientific discovery for all is.

Best known for developing Alpha-Go, the first program to defeat a world champion in the complex game of Go, DeepMind has published more than 1000 research papers – including more than a dozen of nature and science – and several challenging ones. Achieved breakthrough results from AI domain. StarCraft II in protein folding.

DeepMind was founded in London in 2010, and joined forces with Google in 2014 to accelerate its work. Since then, its community has expanded to include teams at Mountain View in Alberta, Montreal, Paris and California.

Leave a Reply