How AlphaFold helps solve one of the foremost challenges in biology today

An experimental biologist and a computational biologist, Ananya Mukherjee and Shweta Ramdas, get together to discuss the new Artificial Intelligence system that can predict the structure of proteins, and its potential relevance to their own work. 

Feat proteinstructure

This year’s Breakthrough Prizes, sometimes dubbed the Oscars of science’, were announced on September 22. With a reward of USD 3 million each, these are currently the most lucrative science prize in the world, topping even the Nobel Prizes! This time, eleven researchers were recognised for their discoveries in fundamental physics, life sciences and mathematics.

Demis Hassabis and John Jumper shared a prize for developing AlphaFold, an Artificial Intelligence (AI) system which accurately predicts the structure of proteins, one of the foremost challenges in biology today. Everything in life is run by proteins — from enzymes to drugs to pathogen proteins,” said Ananya Mukherjee, biologist and faculty member at Azim Premji University. 

While it is now easy to figure out the chain of amino acids that constitute any given protein, it’s more difficult to predict how this chain folds. A single protein can take way more shapes than can be tested, but somehow inside the cell, the protein knows it must take one particular shape and that’s when it can perform its function. So far, we don’t know what rules govern the way a protein folds,” said Shweta Ramdas, a biologist and faculty member at Azim Premji University, summarising the protein structure prediction problem.

It’s all out there. I looked up my favourite proteins and I could vote if I think this is a good structure or not. This is great. Not only can we use the technique, but we can also give our feedback. This is exactly what open science is about, right?”

Researchers led by Hassabis and Jumper made a breakthrough with a machine learning-based system they developed called AlphaFold. AlphaFold recently published the structures of 200 million proteins, that’s nearly every known protein from across the tree of life! As remarkable as this feat itself is the fact that all this information has been made available to the public for free. 

Ananya, an experimental biologist studying algal photosynthesis, appreciated this aspect of AlphaFold: It’s all out there. I looked up my favourite proteins and I could vote if I think this is a good structure or not. This is great. Not only can we use the technique, but we can also give our feedback. This is exactly what open science is about, right?”

The value of possessing knowledge of a protein’s 3D structure is well demonstrated in recent history, with applications ranging from drug discovery to antibiotic resistance, from synthetic biology to crop resilience. Shweta pointed out the most pertinent example of the COVID-19 virus. 

The discovery of the structure of the virus’s spike protein allowed us to understand how it can recognise and bind to a human cell. This allows us to target it,” she said. Ananya added that there was also a group of scientists using protein folding prediction to find plastic-degrading enzymes from the ocean. There’s so much that protein folding can tell us,” she stressed.

Are these crystal structures truly reflective of what is going on inside the cell? Proteins actually look different in the cell. Many function in groups, or they may pair up with a metal ion. Predicting the structure of these complexes is even harder.”

So, does AlphaFold mark the end of the protein structure prediction problem in science? Not really. Shweta reminded us that being a computational system, there are limitations. It’s very good at predicting new protein structures based on what we already know about existing crystal structures. So, your predictions are only as good as the crystal structures are,” she explained. 

There was also something else to consider: Are these crystal structures truly reflective of what is going on inside the cell? Proteins actually look different in the cell. Many function in groups, or they may pair up with a metal ion. Predicting the structure of these complexes is even harder.”

Nevertheless, both the experimental and computational biologists are excited about the potential ways in which AlphaFold can bolster their own areas of research. Ananya’s research explores the possibility of transferring specific algal proteins into food crops to improve their efficiency of photosynthesis. The more we know about these proteins, their X‑ray crystallography structure & their predicted structures, we will be able to see which ones make sense to mutate and transfer to higher plants,” she said. 

Being a computational biologist working on human genetics, Shweta is keen to make the most of AlphaFold for her projects. We have DNA from individuals who have a certain developmental disorder, and we found mutations in a particular gene in these individuals — mutations that have never been found before. We want to predict what these mutations do and if they are causing the disease. Feeding these mutations into AlphaFold can help us answer this.”

Watch Shweta and Ananya discuss this breakthrough discovery:

References

About the Author

Nandita Jayaraj is a Science writer and Communications Consultant at Azim Premji University.

Know more about BSc in Biology at Azim Premji University here.

Attribution