[ad_1]
Thanks to AI, we have just had incredibly powerful tools to decode life.
In two consecutive articles last week, scientists at DeepMind and the University of Washington described deep learning-based methods to solve protein folding – the last step in executing programming in our DNA. and a “one-time lead in a generation”. “
Proteins are the servants of life. They shape our bodies, fuel our metabolism, and are the targets of most drugs today. They start off as a simple ribbon, translated from DNA, then fold into complex three-dimensional architectures. Much like Transformers, many protein units assemble into massive, mobile complexes that change structure according to their current functional needs.
Misfolded proteins can be devastating, causing health problems ranging from sickle cell disease to cancer and Alzheimer’s disease. One of the biggest challenges in biology over the past 50 years has been deciphering how a simple one-dimensional ribbon-like structure transforms into 3D shapes, equipped with canyons, ridges, valleys and caves. It’s like an alien read the coordinates of hundreds of locations on a map of the Grand Canyon on a laptop and reconstruct them into a 3D hologram of the real thing, never seeing it or knowing what it should look like.
Yes. It’s hard. “A lot of people have broken their heads,” said Dr. John Moult of the University of Maryland.
It is not just an academic exercise. The resolution of the human genome paved the way for gene therapy, CAR-T breakthroughs in cancer, and the infamous CRISPR gene editing tool. Deciphering protein folding is meant to illuminate a whole new landscape of biology that we haven’t been able to study or manipulate. The rapid and furious development of the Covid-19 vaccines has relied on scientists analyzing several protein targets on the virus, including the advanced proteins that the vaccines target. Many proteins that lead to cancer are so far beyond the reach of drugs because their structure is difficult to pinpoint.
With these new AI tools, scientists could solve haunting medical mysteries while preparing to face those as yet unknown. It paves the way for a better understanding of our biology, information about new drugs and even inspiration from synthetic biology down the line.
“What the DeepMind team has been able to achieve is fantastic and will change the future of structural biology and protein research,” said Dr Janet Thornton, Director Emeritus of the European Institute of Bioinformatics.
“I never thought I would see this in my lifetime,” Moult added.
Birth of a protein
Imagine life as a video game. If DNA is the basic background code, then proteins are its execution – the actual game you are playing. Any bug in the DNA could cause the program to crash, but it could also be mild and allow the game to work as usual. In other words, most modern medicines, like gamers, only care about the end gameplay – the proteins – rather than the source code that leads to it, unless something goes wrong. From diabetes drugs to antidepressants and senolytics that potentially extend life, these drugs all work by clinging to proteins rather than DNA.
This is why deciphering the structure of proteins is so important: Like a key to a lock, a drug can only anchor itself to a protein in specific places. Likewise, proteins often associate by binding together in a complex to perform functions in your body, such as forming a memory or triggering an immune attack against a virus.
Proteins are made up of building blocks called amino acids, which are in turn programmed by DNA. Similar to Rosetta Stone, our cells can easily translate the DNA code into building blocks of proteins inside a clam shell-like structure, which spits out a one-dimensional chain of amino acids. These ribbons are then mixed through an entire cellular infrastructure that allows the protein to fold into its final structure.
In the 1970s, Nobel Laureate Dr Christian Anfinsen claimed that the one-dimensional sequence itself can computer predict the 3D structure of a protein. The problem is time and power: like trying to crack a password with hundreds of characters suspended in 3D space, the potential solutions are astronomical.
But now we have a tool that beats humans at finding patterns: machine learning.
Enter AI
In 2020, DeepMind shocked the entire domain with its entry into a former biennial competition. Dubbed CASP (Critical Assessment of Protein Structure Prediction), the decades-long test uses traditional laboratory methods to determine protein structure as the basis for judging prediction algorithms.
The baseline is hard to come by. It relies on laborious experimental techniques that can take months or even years. These methods often “freeze” a protein and map its internal structure down to the atomic level using x-rays. Many proteins cannot be processed in this way without losing their natural structure, but the method is best. we currently have. The predictions are then compared to this gold standard to judge the underlying algorithm.
Last year, DeepMind stunned everyone with its AI, blasting the other contestants out of the water. At the time, they were a tease, revealing few details about their “insanely exciting” method that matched the experimental results precisely. But the 30-minute presentation inspired Dr Minkyung Baek of the University of Washington to develop his own approach.
Baek used a similar deep learning strategy, described in an article by Science this week. The tool, RoseTTAFold, simultaneously considers three levels of models. The first examines the building blocks of amino acids in a protein and compares them to all the other sequences in a protein database.
The tool then examines how amino acids from one protein interact with another within the same protein, for example, by examining the distance between two distant building blocks. It’s like looking at your hands and feet fully outstretched from a backbend, and measuring the distance between those ends when you “tuck in” into a yoga pose.
Finally, the third track examines the 3D coordinates of each atom that makes up a protein building block – much like mapping plots on a Lego block – to compile the final 3D structure. The network then bounces between those tracks, so that one output can update another track.
The final results were close to those of DeepMind’s tool, AlphaFold2, which corresponded to the gold standard of structures obtained from the experiments. Although RoseTTAFold was not as accurate as AlphaFold2, it apparently took a lot less time and energy. For a simple protein, the algorithm was able to solve the structure using a gaming computer in about 10 minutes.
RoseTTAFold was also able to solve the problem of “protein assembly”, in that it could predict the structure of proteins, made up of several units, by simply examining the amino acid sequence alone. For example, they were able to predict how the structure of an immune molecule locks onto its target. Many biological functions rely on these handshakes between proteins. Being able to predict them using an algorithm opens the door to the manipulation of biological processes (immune system, stroke, cancer, brain function) that we did not have access to before.
Body hacking
Since RoseTTAFold’s public release in July, it has been downloaded hundreds of times, allowing other researchers to answer their bewildering protein sequence questions, potentially saving years of work while collectively improving the algorithm.
“When there is a breakthrough like this two years later, everyone is doing it as well if not better than before,” said Moult.
Meanwhile, DeepMind is also releasing its AlphaFold2 code, the one that inspired Baek.
In a new paper in Nature, the DeepMind team described their approach to the 50-year-old mystery. The key was to integrate multiple sources of information – the evolution of a protein and its physical and geometric constraints – to build a two-step system that maps a given protein with surprisingly high precision.
First presented at the CASP meeting, Dr Demis Hassabis, founder and CEO of DeepMind, is ready to share the code with the world. “We are committed to sharing our methods and providing broad and free access to the scientific community. Today, we are taking the first step towards realizing that commitment by sharing AlphaFold’s open source code and releasing the complete system methodology, ”he wrote, adding that“ we are delighted to see which other new avenues of research this will help the community. “
With the two studies, we enter a new world of predicting, and then engineering or modifying, the building blocks of life. Dr Andrei Lupas, evolutionary biologist at the Max Planck Institute for Developmental Biology and CASP judge, agrees: “This is going to change medicine. It’s going to change research, ”he said. “It will change bioengineering. It will change everything. “
Image Credit: Ian Haydon, University of Washington Protein Design Institute
[ad_2]
Source link