Updated: Mar 31
DeepMind has made significant advances in figuring out what shapes proteins fold into. One would ask, why should we be concerned about folding patterns of proteins? Proteins, that are made up of chains of amino acids, are practically essential to all of life’s functions. A protein's function is highly dependent on its 3D structure. However, determining what shapes a protein folds into has been one of biology’s greatest challenges and scientists and biologists have been perplexed by this problem for a long time.
Recently, the organizers of the Critical Assessment of protein Structure Prediction (CASP) recognized DeepMind’s AlphaFold 2 to be a possible solution to this problem. This can prove to be groundbreaking as a protein’s function is dependent on its structure and in-depth knowledge could help speed up the process of vaccine development. DeepMind used AlphaMind to study a variety of proteins that constitute the virus causing Covid-19.
The CASP14 Results
CASP14 measures the accuracy of folding pattern predictions using the Global Distant Test (GDT) which ranges from 0-100. A GDT score of 90 is considered to be competitive with experimental methods. The AlphaFold 2 system had a median score of 92.4 GDT.
AlphaMind uses advanced deep learning architecture to predict the folding patterns of proteins. They used publicly available data of ~170,000 proteins and their physical structures to train the model so that it can predict the shapes of proteins with high accuracy.
Studying the Covid-19 Virus:
Covid-19 is caused by the SARS-CoV-2 virus. 30 different proteins constitute this virus and there was limited understanding about 10 of those proteins. DeepMind’s algorithm also accurately predicted the spike structure of the SARS-CoV-2 virus that has already been experimentally determined. This speaks to the potential of this algorithm and how it could be revolutionary in determining the structure of various proteins. DeepMind used AlphaFold to study the proteins in the SARS-CoV-2 virus and predict their structures. They demonstrated highly accurate predictions for the ORF8 protein that is found in the SARS-CoV-2 virus.
Future Potential/Bridge Point of View
Since proteins are responsible for a large percentage of life’s functions, the solution to the world's greatest problems (vaccine development, waste management) lies in predicting the shape of proteins. Such advancements in this field could help us better prepare for future pandemics and find solutions at a much faster pace.
John Jumper, Kathryn Tunyasuvunakool, Pushmeet Kohli, Demis Hassabis, and the AlphaFold Team, “Computational predictions of protein structures associated with COVID-19”, Version 3, DeepMind website, 4 August 2020, https://deepmind.com/research/open-source/computational-predictions-of-protein-structures-associated-with-COVID-19
John Jumper, Kathryn Tunyasuvunakool, Pushmeet Kohli, Demis Hassabis, and the AlphaFold Team, “High Accuracy Protein Structure Prediction Using Deep Learning”, DeepMind website, 30 November 2020, https://deepmind.com/blog/article/alphafold-a-solution-to-a-50-year-old-grand-challenge-in-biology