April 15, 2024
A.I

AlphaFold found thousands of potential psychedelics. Will their predictions help drug discovery?

Protein structures predicted by AlphaFold have helped identify candidate drug compounds.Credit: Deep Mind

Researchers have used the protein structure prediction tool AlphaFold to identify1 hundreds of thousands of potential new psychedelic molecules, which could help develop new types of antidepressants. The research shows, for the first time, that AlphaFold predictions, available at the touch of a button, can be as useful for drug discovery as experimentally derived protein structures, which can take months or even years to determine.

The development is a boost for AlphaFold, the artificial intelligence (AI) tool developed by DeepMind in London that has been a game-changer in biology. The public AlphaFold database contains structural predictions for almost all known proteins. The protein structures of molecules implicated in diseases are used in the pharmaceutical industry to identify and improve promising drugs. But some scientists had begun to doubt whether AlphaFold’s predictions could replace standard experimental models in the search for new drugs.

“AlphaFold is an absolute revolution. “If we have a good structure, we should be able to use it to design drugs,” says Jens Carlsson, a computational chemist at Uppsala University in Sweden.

AlphaFold Skepticism

Efforts to apply AlphaFold to find new drugs have been met with considerable skepticism, says Brian Shoichet, a pharmaceutical chemist at the University of California, San Francisco. “There is a lot of commotion. “Anytime someone says ‘so-and-so is going to revolutionize drug discovery,’ it deserves some skepticism.”

Shoichet counts more than ten studies that have found that AlphaFold predictions are less useful than protein structures obtained with experimental methods, such as X-ray crystallography, when used to identify potential drugs in a modeling method called protein-docking. ligand.

This approach, common in the early stages of drug discovery, involves modeling how hundreds of millions or billions of chemicals interact with key regions of a target protein, in the hope of identifying compounds that alter the protein’s activity. Previous studies have tended to find that when using structures predicted by AlphaFold, the models are poor at selecting drugs that are already known to bind to a particular protein.

Researchers led by Shoichet and Bryan Roth, a structural biologist at the University of North Carolina at Chapel Hill, reached a similar conclusion when they compared the AlphaFold structures of two proteins implicated in neuropsychiatric conditions with known drugs. The researchers wondered whether small differences from the experimental structures could cause the predicted structures to miss certain protein-binding compounds, but also allow them to identify others that were no less promising.

To test this idea, the team used experimental structures of the two proteins to screen virtually hundreds of millions of potential drugs. One protein, a receptor that senses the neurotransmitter serotonin, was previously determined using cryo-electron microscopy. The structure of the other protein, called σ-2 receptor, had been mapped by X-ray crystallography.

Drug differences

They ran the same screen with protein models drawn from the AlphaFold database. They then synthesized hundreds of the most promising compounds identified with the experimental and predicted structures and measured their activity in the laboratory.

Screening tests with experimental and planned structures yielded completely different drug candidates. “No two molecules were the same,” says Shoichet. “They didn’t even look like each other.”

But to the team’s surprise, the “hit rates” (the proportion of labeled compounds that actually altered protein activity significantly) were almost identical for the two groups. And the AlphaFold structures identified the drugs that activated the serotonin receptor most potently. The psychedelic drug LSD acts in part through this pathway, and many researchers are looking for non-hallucinogenic compounds that do the same, as potential antidepressants. “It’s a really new result,” says Shoichet.

Predictive power

In unpublished work, Carlsson’s team found that AlphaFold structures are good at identifying drugs for a highly sought-after class of targets called G protein-coupled receptors, for which their hit rate is around 60%.

Having confidence in the predicted protein structures could be a game-changer for drug discovery, Carlsson says. Determining structures experimentally is not trivial, and many potential targets may not yield to existing experimental tools. “It would be very convenient if we could push the button and get a structure that we can use for ligand discovery,” he says.

Photo illustration of the Isomorphic Labs logo displayed on a tablet.

Isomorphic Labs, a spin-off of Google’s DeepMind in London, is ramping up its drug discovery efforts using AlphaFold.Credit: Igor Golovniov/SOPA Images/LightRocket via Getty

The two proteins that Shoichet and Roth’s team chose are good candidates to rely on AlphaFold, says Sriram Subramaniam, a structural biologist at the University of British Columbia in Vancouver, Canada. Experimental models of related proteins are readily available, including detailed maps of the regions where drugs bind to them. “If we put it all together, AlphaFold is a paradigm shift. It changes the way we do things,” he adds.

“This is not a panacea,” says Karen Akinsanya, president of therapeutics research and development at Schrödinger, a New York City-based pharmaceutical software company that uses AlphaFold. The predicted structures are useful for some drug targets but not others, and it is not always clear which one applies. In about 10% of cases, predictions that AlphaFold considers very accurate are substantially different from the experimental setup, according to one study.2 found.

And even when predicted structures can help identify clues, more detailed experimental models are often needed to optimize the properties of a particular drug candidate, Akinsanya adds.

Big bet

Shoichet agrees that AlphaFold’s predictions are not universally useful. “There were a lot of models that we didn’t even try because we thought they were too bad,” he says. But he estimates that in about a third of cases, an AlphaFold structure could boost a project. “Compared to going out and getting a new structure, the project could move forward a couple of years and that’s huge,” he says.

That’s the goal of Isomorphic Labs, DeepMind’s drug discovery subsidiary in London. On January 7, the company announced deals worth at least $82.5 million (and up to $2.9 billion if commercial targets are met) to pursue drugs on behalf of pharmaceutical giants Novartis and Eli Lilly using research tools. machine learning like AlphaFold.

The company says the work will be aided by a new version of AlphaFold that can predict the structures of proteins when they bind to drugs and other interacting molecules. DeepMind has not yet said when, or if, the update will be available to researchers, as previous versions of AlphaFold have been. A competing tool called RoseTTAFold All-Atom3 its developers will make it available soon.

Such tools will not completely replace experiments, scientists say, but their potential to help find new drugs should not be discounted. “There are a lot of people who want AlphaFold to do everything, and a lot of structural biologists want to find reasons to say that we are still needed,” says Carlsson. “Finding the right balance is difficult.”

Leave a Reply

Your email address will not be published. Required fields are marked *