Take away food
-
A “deep learning” software program from the Google-owned DeepMind lab has shown great progress in solving one of biology’s biggest challenges – understanding protein folding.
-
Protein folding is the process by which a protein takes its shape from a chain of building blocks to its final three-dimensional structure, which determines its function.
-
By better predicting how proteins adopt their structure, or “fold up”, scientists can more quickly develop drugs that, for example, block the action of crucial viral proteins.
Solving what biologists call “the protein folding problem” is a big deal. Proteins are the workhorses of cells and are present in all living organisms. They are made up of long chains of amino acids and are vital for the structure of cells and the communication between them as well as for regulating all body chemistry.
This week, the Google-owned artificial intelligence company DeepMind demonstrated a deep learning program called AlphaFold2that experts call breakthrough towards the resolution of the great challenge of protein folding.
Proteins are long chains of amino acids linked together like beads on a string. But for a protein to do its job in the cell, it must “bend” – a process of twisting and bending that transforms the molecule into a complex three-dimensional structure that can interact with its target in the cell. If the folding is interrupted, the protein will not take the right shape and will not be able to do its job inside the body. This can lead to disease, as in a common disease like Alzheimer’s disease, and in rare cases like cystic fibrosis.
Deep learning is a computational technique that uses the often hidden information contained in large datasets to solve questions of interest. It has been widely used in fields such as games, speech and voice recognition, self-driving cars, science and medicine.
I think tools like AlphaFold2 will help scientists design new types of proteins, ones that could, for example, help break down plastics and fight future pandemics and viral diseases.
I am a computational chemist and author of the book The state of science. My students and I study the structure and properties of fluorescent proteins using protein folding computer programs based on classical physics.
After decades of study by thousands of research groups, these protein folding prediction programs are very good at calculating the structural changes that occur when we make small changes to known molecules.
But they weren’t good enough at predicting how proteins fold from scratch. Before the advent of deep learning, the problem of protein folding seemed incredibly difficult, and it looked set to frustrate computational chemists for many decades to come.
Protein folding
The amino acid sequence – which is encoded in DNA – defines the 3D shape of the protein. Form determines function. If the structure of the protein changes, it is unable to perform its function. Correctly predicting protein folds based on amino acid sequence could revolutionize drug design and explain the causes of new and old diseases.
All proteins with the same sequence of amino acid building blocks fold into the same three-dimensional shape, which optimizes amino acid interactions. They do this in milliseconds, although they have an astronomical number of possible configurations – about 10 power 300. This huge number is what makes it difficult to predict how a protein folds even when scientists know the full sequence of amino acids that go into making it. Previously, it was impossible to predict protein structure from amino acid sequence. Protein structures have been determined experimentally, a time-consuming and expensive undertaking.
Once researchers can better predict how proteins fold, they can better understand how cells work and how misfolded proteins cause disease. Better protein prediction tools will also help us design drugs that can target a particular topological region of a protein where chemical reactions are taking place.
AlphaFold was born from the deep learning of Chess, Go and Poker games
The success of DeepMind’s protein folding prediction program, called Alpha folding, is not unexpected. Other deep learning programs written by DeepMind demolished the best chess, go and poker players in the world.
In 2016 Stockfish-8, an open-source chess engine, was the world champion in computer chess. It was evaluating 70 million chess positions per second and had centuries of accumulated human chess strategies and decades of computer experience to draw upon. He played efficiently and brutally, ruthlessly defeating all his human challengers without an ounce of finesse. Enter deep learning.
On December 7, 2017, Google’s Deep Learning Chess Program AlphaZero beat Stockfish-8. Chess engines have played 100 games, with AlphaZero winning 28 and tying 72. He hasn’t lost a single game. AlphaZero performed only 80,000 calculations per second, compared to Stockfish-8’s 70 million calculations, and it only took four hours to learn chess from scratch by playing against itself a few million times and by optimizing its neural networks as it learns.
AlphaZero learned nothing from humans or chess games played by humans. He taught himself and in the process derived strategies never seen before. In a comment in Science magazine, former world chess champion Garry Kasparov wrote that by learning by playing himself, AlphaZero developed strategies that “reflect the truth” of chess rather than reflect “the priorities and prejudices” of programmers. “It’s the epitome of the ‘work smarter, not harder’ cliché.”
CASP – the Molecular Modellers Olympics
Every two years, the world’s top computational chemists test their programs’ abilities to predict protein folding and compete in the Critical evaluation of structure prediction (CASP) competition.
In the competition, teams receive the linear amino acid sequence of approximately 100 proteins whose 3D shape is known but not yet published; they must then calculate how these sequences would fold. In 2018, AlphaFold, the competition’s deep learning rookie, beat all traditional programs – but barely.
Two years later, on Monday, it was announced that Alphafold2 had won the 2020 contest by a healthy margin. It beat its competitors and its predictions were comparable to existing experimental results determined by gold standard techniques such as X-ray diffraction crystallography and cryo-electron microscopy. Soon, I expect AlphaFold2 and its offspring to be the methods of choice for determining protein structures before resorting to experimental techniques that require painstaking and laborious work on expensive instrumentation.
One of the reasons for AlphaFold2’s success is that it could use the Protein databasewhich has more than 170,000 experimentally determined 3D structures, to practice calculating the correctly folded structures of proteins.
The potential impact of AlphaFold can be appreciated by comparing the number of all published protein structures – approximately 170,000 – with the 180 million DNA and protein sequences deposited in the Universal Protein Database. AlphaFold will help us sift through treasure troves of DNA sequences in search of new proteins with unique structures and functions.
Has AlphaFold made me, a molecular modeler, redundant?
As with the Chess and Go programs – AlphaZero and AlphaGo – we don’t know exactly what the AlphaFold2 algorithm does and why it uses certain correlations, but we do know that it works.
In addition to helping us predict important protein structures, understanding AlphaFold’s “reflection” will also help us gain new insights into the mechanism of protein folding.
[Deep knowledge, daily. Sign up for The Conversation’s newsletter.]
One of the most common fears expressed about AI is that it will lead to large-scale unemployment. AlphaFold still has a long way to go before it can consistently and successfully predict protein folding.
However, once it has matured and the program can simulate protein folding, computational chemists will be integrally involved in improving the programs, trying to understand the underlying correlations used and applying the program to solve important problems such as protein misfolding associated with many diseases. such as Alzheimer’s disease, Parkinson’s disease, cystic fibrosis and Huntington’s disease.
AlphaFold and its offspring will certainly change the way computational chemists work, but it won’t make them redundant. Other regions will not be so lucky. In the past, robots could replace humans doing manual labor; with AI, our cognitive skills are also challenged.
