Tech

AI predicts the shape of the coming molecule

Biologist John McGeehan, director of the Enzyme Innovation Center in Portsmouth, UK, has been looking for molecules that can break down 150 million tonnes of soda bottles and other plastic wastes scattered around the world over the last few years. Was there.

In collaboration with researchers on both sides of the Atlantic, he discovered Some good options.. But his job is the most demanding locksmith’s job. It is to identify compounds that themselves twist and fold, fit perfectly into the molecules of PET bottles, and fold into fine shapes that can split them like a key to open a door. ..

Determining the exact chemical content of a given enzyme is a fairly simple challenge these days. However, years of biochemical experiments may be required to identify its three-dimensional shape.Last fall, the Institute for Artificial Intelligence in London DeepMind has built a system that automatically predicts the shape of enzymes and other proteins.Dr. McGeehan asked the lab if it would be useful for his project.

Towards the end of the week, he sent DeepMind a list of seven enzymes. Next Monday, the lab returned all seven shapes. “This moved us a year ago, if not two,” said Dr. McGeehan.

Now any biochemist can speed up work in much the same way. On Thursday, DeepMind announced the predicted shape of over 350,000 proteins. This is a microscopic mechanism that drives the behavior of bacteria, viruses, the human body, and all other organisms. This new database contains the 3D structures of all proteins expressed by the human genome and the 3D structures of proteins that appear in 20 other organisms such as mice, fruit flies, and E. coli.

This vast and detailed biological map offers about 250,000 previously unknown shapes that could accelerate the ability to understand diseases, develop new drugs, and reuse existing drugs. There is. It can also lead to new types of biological tools, such as enzymes that efficiently break down PET bottles and convert them into materials that are easy to reuse and recycle.

Gira Bhabha, an assistant professor of cell biology at New York University, said: “Whatever the field of biology, whether studying neuroscience or immunology, this can be useful.”

This new knowledge is the key of its own kind. If scientists can determine the shape of a protein, they can determine how other molecules bind to it. This may, for example, reveal how bacteria resist antibiotics and how to counter them. Bacteria resist antibiotics by expressing certain proteins. If scientists can identify the shape of these proteins, they can develop new antibiotics and new drugs to suppress them.

In the past, accurate identification of protein shapes required months, years, and even decades of trial and error experiments using X-rays, microscopes, and other tools on the bench. However, DeepMind can significantly shorten its timeline with AI technology known as AlphaFold.

When Dr. McGeehan sent DeepMind a list of seven enzymes, he told the lab that he had already identified two of them, but didn’t say which one. This was a way to test how well the system was working. AlphaFold passed the test and correctly predicted both shapes.

Dr. McGeehan said it was even more noteworthy that the forecasts arrived within a few days. He later learned that AlphaFold actually completed the task in just a few hours.

AlphaFold is so-called neural networkA mathematical system that can learn tasks by analyzing vast amounts of data (in this case thousands of known proteins and their physical shapes) and extrapolating them to the unknown.

this is, Identifies commands that bark on your smartphone, Recognize faces in photos posted on Facebook And that Translate one language into another With Google Translate and other services. However, many experts believe that AlphaFold is one of the most powerful applications of technology.

“This shows that AI can do useful things in the complexity of the real world,” said one of the creators of the AI ​​Index, an effort to track advances in artificial intelligence technology around the world. Jack Clark says.

As Dr. McGeehan discovered, it can be very accurate. AlphaFold has an approximately 63% chance of predicting protein shape with accuracy comparable to physics experiments, according to independent benchmark tests that compare known protein structures with predictions. Most experts thought that such a powerful technology was still years away.

Professor Randy Reed of the University of Cambridge said: “This was a complete change.”

However, due to the varying accuracy of the system, some DeepMind database predictions are less useful than others. Each prediction in the database comes with a “confidence score” that indicates how likely it is to be accurate. DeepMind researchers estimate that the system provides “good” predictions in about 95% of the time.

As a result, the system cannot completely replace physics experiments. It is used in conjunction with bench work to help scientists determine which experiments need to be performed and fill gaps when an experiment fails. Researchers at the University of Colorado at Boulder have recently used AlphaFold to help identify protein structures that have been difficult to identify for over a decade.

The developers of DeepMind chose to freely share a database of protein structures rather than selling access, hoping to spur progress in bioscience as a whole. “We are interested in the biggest impact,” said DeepMind, CEO and co-founder of DeepMind, which is owned by the same parent company as Google but operates more like a laboratory than a for-profit business. One Demis Hassabis says.

Some scientists have compared DeepMind’s new database with the Human Genome Project. The Human Genome Project, completed in 2003, provided a map of all human genes. DeepMind currently provides a map of approximately 20,000 proteins expressed by the human genome. This is another step in understanding how our body works and how we can respond when problems arise.

It is also expected that technology will continue to evolve. The University of Washington lab has built a similar system called RoseTTAFold and, like DeepMind, is openly sharing the computer code that drives the system. Anyone can use this technology and anyone can work to improve it.

Even before DeepMind began sharing technology and data openly, AlphaFold provided information for a wide range of projects. Researchers at the University of Colorado are using this technology to understand how bacteria such as Escherichia coli and Salmonella develop resistance to antibiotics and to develop ways to counter this resistance. At the University of California, San Francisco, researchers are using this tool to better understand the coronavirus.

Coronavirus causes havoc in the body through 26 proteins. With the help of AlphaFold, researchers Improved understanding of one important protein And we hope that technology will help us to better understand the other 25.

If this is too late to affect the current pandemic, it may help prepare for the next pandemic. “A better understanding of these proteins will help target this virus as well as other viruses,” said Kliment Verba, one of the researchers in San Francisco.

There are countless possibilities. After DeepMind gave Dr. McGeehan the shape of seven enzymes that could get rid of the world of plastic waste, he sent an additional 93 lists to the lab. “They are working on these now,” he said.

AI predicts the shape of the coming molecule

Source link AI predicts the shape of the coming molecule

Back to top button