Why DeepMind’s AlphaFold 2 is a breakthrough in biology?

On 30th November 2020, DeepMind’s program, AlphaFold, was named the best program in the Critical Assessment of Structure Prediction (CASP). CASP is a challenge that is held every two years to predict the protein-structures from amino-acid sequences and more than 100 research groups from around the world participate. DeepMind first participated in CASP in 2018, and their initial version of AlphaFold secured the top spot. But in 2020, DeepMind was able to achieve startling accuracy. So, why is this important?

A score greater than 90 is considered to be competitive with experimental results. [1]

First, we need to understand why predicting protein-structures is essential.

What is protein folding, why is it important and how AlphaFold offers a solution?

The human body is composed of trillions of cells, and each cell contains proteins that play critical roles. Proteins support not just the biological processes in your body but every biological process in living things. They are the building blocks of life. [1] The structure of each protein determines its function. For example — Hemoglobin (a protein) is responsible for transporting oxygen in the blood of vertebrates. This protein consists of 4 chains (2 α chains and 2 β chains). If there is a substitution in the β chain, it will cause a change in the protein structure, leading to sickle cell disease. [2]

Sickle cells are crescent shaped, while normal cells are disc-shaped. [2]

Therefore, understanding the structure of proteins in diseases can help us develop new drugs that can save millions of lives.

As of 1st April 2020, about 170,000 protein structures are known. [3] These structures were determined using experimental methods such as X-ray crystallography, NMR spectroscopy, and electron microscopy. [4] So if we do have ways to determine protein structures, why do we need a program like AlphaFold?

Experimental methods take a long time (sometimes years) to produce results, and often the structures are determined through comprehensive trial and error. [5] Whereas the use of computational methods can significantly speed up the discovery process with high accuracy. Professor Andrei Lupas, Director at the Max Planck Insitute of Developmental Biology in Tübingen, Germany, has spent over a decade trying to determine a bacterial protein structure. With DeepMind’s model, his lab was able to determine the structure in half an hour. [6]

AlphaFold’s high accuracy and speed will transform biology. It will help the research community understand diseases better, accelerate drug discovery, and potentially find new methods to break down industrial and plastic waste. [1]

References

1. High Accuracy Protein Structure Prediction Using Deep Learning, John Jumper, Richard Evans, Alexander Pritzel, Tim Green, Michael Figurnov, Kathryn Tunyasuvunakool, Olaf Ronneberger, Russ Bates, Augustin Žídek, Alex Bridgland, Clemens Meyer, Simon A A Kohl, Anna Potapenko, Andrew J Ballard, Andrew Cowie, Bernardino Romera-Paredes, Stanislav Nikolov, Rishub Jain, Jonas Adler, Trevor Back, Stig Petersen, David Reiman, Martin Steinegger, Michalina Pacholska, David Silver, Oriol Vinyals, Andrew W Senior, Koray Kavukcuoglu, Pushmeet Kohli, Demis Hassabis. In Fourteenth Critical Assessment of Techniques for Protein Structure Prediction (Abstract Book), 30 November — 4 December 2020. Retrieved from here.
2. Protein Structure. Lumen. https://courses.lumenlearning.com/boundless-chemistry/chapter/protein-structure/
3. Protein Data Bank. https://en.wikipedia.org/wiki/Protein_Data_Bank
4. Methods for Determining Atomic Structures. https://pdb101.rcsb.org/learn/guide-to-understanding-pdb-data/methods-for-determining-structure
5. Understanding Crystallography — Part 1: From Proteins to Crystals. https://www.youtube.com/watch?v=gLsC4wlrR2A
6. ‘It will change everything’: DeepMind’s AI makes gigantic leap in solving protein structures, Ewen Callaway. Nature 588, 203–204 (2020). https://www.nature.com/articles/d41586-020-03348-4