Advertisement

technologyTechnology
clockPUBLISHED

DeepMind's New AI May Have Just Solved "The Greatest Challenge In Biology"

author

Jack Dunhill

author

Jack Dunhill

Social Media Coordinator and Staff Writer

Jack is a Social Media Coordinator and Staff Writer for IFLScience, with a degree in Medical Genetics specializing in Immunology.

Social Media Coordinator and Staff Writer

3D illustration of a concept false-color protein. Christoph Burgstedt / Shutterstock.com

When it comes to life, proteins are everything. The production of proteins from genes underpins every cellular process, every difference in how a person looks, every movement you make. Producing these proteins relies on a complex system of folding amino acids (the building blocks produced by our genetic code) over and over to create intricate structures that determine how the protein will act and what it will act upon. Despite massive leaps and technological advancements in the study of proteins, understanding how proteins fold and what shape a simple amino acid code will produce has eluded scientists. This is called the “protein folding problem”, and is one of the greatest challenges of biology.

However, in a breakthrough by UK-based artificial intelligence company DeepMind's AlphaFold team, scientists believe they have found the solution in AI. Hailed as an achievement that will “transform biology and medicine,” the deep-learning system may be able to simulate protein structures from just an amino acid code, a feat that usually takes entire PhDs to complete.

Advertisement

“We have been stuck on this one problem – how do proteins fold up – for nearly 50 years. To see DeepMind produce a solution for this, having worked personally on this problem for so long and after so many stops and starts, wondering if we’d ever get there, is a very special moment,” said Professor John Moult, co-founder and chair of the Critical Assessment of protein Structure Prediction (CASP), in a statement.

Protein structures are notoriously difficult to figure out. Our current methods include X-ray crystallography, which involves crystallizing the protein sample before X-ray imaging it and compiling the electron density data to create a 3D structure, or cryo-electron microscopy, which freezes samples to cryogenic temperatures before 3D imaging. These have given us impressive insight into protein structures, but some proteins cannot be imaged in this way, and both take large amounts of time and are incredibly expensive.

Alongside this, both techniques will never solve the protein folding problem, as they only image the samples presented before them – what if you want to predict a protein structure from its amino acid sequence?

Instead, researchers tried a different approach – they created an online game for people around the globe to participate in. The game, called Foldit, was a crowdsourced effort to predict protein folding by allowing users to predict their own protein shape for a given sequence, with the highest scoring model winning. As innovative an approach as this is, it is time-consuming, laborious and often inaccurate.

Advertisement

In an attempt to solve the problem, DeepMind recruited artificial intelligence to do what mere mortals cannot. Using deep-learning, they created an AI-driven system that can predict protein structures from basic amino acid sequences to an incredible degree of accuracy in a comparatively short time of just a few days.

"We trained this system on publicly available data consisting of ~170,000 protein structures from the protein data bank together with large databases containing protein sequences of unknown structure," state the AlphaFold developers. "It uses approximately 128 TPUv3 cores (roughly equivalent to ~100-200 graphic processing units) run over a few weeks, which is a relatively modest amount of compute in the context of most large state-of-the-art models used in machine learning today."

Two examples of protein targets in the free modeling category. AlphaFold predicts highly accurate structures when measured against experimental results. Credit: DeepMind

Whilst the official data has not been published yet, the announcement has left the scientific community at peak excitement and speculation of what this will mean for structural biology. A full understanding of protein folding would leap fields such as medicine forward, possibly enabling more effective and more tailored drugs to be produced at a far faster rate than ever before.

“This computational work represents a stunning advance on the protein-folding problem, a 50-year-old grand challenge in biology. It has occurred decades before many people in the field would have predicted. It will be exciting to see the many ways in which it will fundamentally change biological research,” said Professor Venki Ramakrishnan, Nobel Laureate and President of the Royal Society.


ARTICLE POSTED IN

technologyTechnology
FOLLOW ONNEWSGoogele News