The greatest success of Artificial Intelligence in Life Sciences

Science

28 december 2020
Article
“Overwhelming!”, “A significant breakthrough!”, “A triumph!” are some of the comments from the scientific community regarding AlphaFold’s ability to predict protein structures.

by Maria Giulia Manzione

Contributing Writer

Artificial Intelligence (AI) refers to “the ability of a computer or machine to mimic the capabilities of the human mind”. The concept of AI was introduced in 1950 by Alan Turing in the article Computing Machinery and Intelligence where he described a test (now known as the Turing Test) to determine if a computer is able to think as a human being. AI means creating algorithms — a list of instructions and rules that a computer needs to follow in order to solve a problem — to make decisions and draw predictions from data. But that is not a passive process, as AI algorithms can learn from new data and improve over time. The progress of AI’s applications in our lives, from social media to self-driving cars, has been unprecedented.

In 2010, a team of scientists and engineers as well as machine learning and policy experts founded DeepMind, the most famous AI company developing AI systems that are able to predict eye diseases from routine scans, reduce energy use for data centre cooling, improve accuracy and efficiency of breast cancer screening and much more. However, the most exciting achievement of DeepMind’s team has been the recently improved version of AlphaFold(AlphaFold2), an artificial intelligence algorithm able to predict protein structures.

Proteins are essential elements of all living cells. In humans, all biological, neurological, and metabolic functions involve countless proteins. All proteins are made of amino acids that, during protein synthesis, initially form a single chain, called a polypeptide. Then, in 85% of the cases, the amino acids chain folds into a unique three-dimensional (3D) structure that determines how the protein will function and, sometimes, where the protein will be localized in the cell. Mis-folded proteins are the cause of severe diseases associated with improper cellular localization of the protein (e.g., cystic fibrosis), toxic protein aggregates (e.g., Alzheimer’s disease), and cell shape alteration (e.g., sickle cell anemia). Therefore, knowing the structure of a protein is extremely important not only to understand and study the biological function of the protein but also to develop drugs that can restore the protein’s structure.

During the past 50 years, researchers have been struggling to solve the 3D structure of proteins using expensive and time-consuming techniques such as X-ray crystallography — a method to determine the three-dimensional structure of proteins from the diffraction pattern of electromagnetic radiation (X-rays) passing through crystals of the protein. That process can take up to 1 year and costs more than 100,000 dollars. In contrast, highly advanced computational methods could be helpful to solve protein structures faster and with high accuracy.

In 1994, Professor John Moult and Professor Krzysztof Fidelis founded the Critical Assessment of protein Structure Prediction (CASP), a worldwide biennial competition (like the Olympics of protein folding) where a group of researchers create algorithms to predict the structure of proteins for which another group of scientists determine it experimentally. Then, the “computational” and the “experimental” structures are compared and scored. When the score is close to 90 it means that the prediction is equivalent to the experimentally determined structure. In 2018 (CASP13), and again in 2020 (CASP14), DeepMind’s AlphaFold outperformed with the highest degree of accuracy in predicting the 3D structure of more than 100 proteins. The results of CASP14 are expected to be published in 2021.

Animation from DeepMind's blog post on "AlphaFold: Using AI for scientific discovery"

During the past summer, the team of DeepMind has improved AlphaFold algorithms even further to face the challenges of solving SARS-CoV-2 protein structures. DeepMind has recently announced that AlphaFold could give an accurate prediction for the experimentally determined SARS-CoV-2 spike protein structure. And more will come.

AI emerges as a fast and cost-effective approach to solve the structure of a thousand proteins. It will be one of the most powerful tools for scientists and it will accelerate drug discovery to tackle severe diseases, including COVID-19. Some already predict (and hope) that progress in AI might lead to a Nobel Prize in the future.

For more regular content

For submissions or applications

  • Or message us on Facebook