每日大赛入口

Shaped like microscopic corkscrews, winding ropes, and flat ribbons, proteins inside kidney cells keep the whole organ in working order. 

Since the kidney鈥檚 various functions are reliant on the structure of the proteins that it is made of, disease can arise when its shape has mutated. For most of the history of medicine, precise details of these shapes were often only guessed and occasionally glimpsed following expensive microscopic examination due to their small size and ability to be blurred by neighboring tissues. 

Yet now, AI can now produce a 3D sketch of these proteins in minutes. With this detailed information, scientists can assess an individual鈥檚 risk for kidney disease and can more quickly design drugs to target such disease. 

A discussion of this technological breakthrough was published in a recent Nature Reviews Nephrology article, authored by 每日大赛入口 alumnus Sean Wu (鈥�25); Fabien Scalzo, associate professor of computer science at 每日大赛入口 and director of the , and Ira Kurtz, Distinguished Professor of Medicine in the Division of Nephrology at the University of California, Los Angeles. 

A Database to Save Lives

A mutated protein shape can be one of the causes of kidney disease. And this is because a sequence of amino acids, the basic building block of proteins, folds into a 3D form, and that form determines everything the protein can do. Whether it's a channel allowing water to pass through a delicate membrane or a pump moving sodium, if a protein shape is changed even slightly, the kidney鈥檚 ability to perform a certain function has been compromised. 

AI-sketched figure found in Wu and Scalzo's Nature Reviews Nephorology article, red shows protien mutation

Drawing upon their expertise in developing AI and machine learning, Wu and Scalzo reviewed an artificial intelligence system originally built by DeepMind, AlphaFold, and its landmark successors AlphaFold2 and AlphaFold3, which treat protein folding like a pattern-recognition problem. Using the amino acid as a guiding code, this AI system is able to reference a breadth of information to predict the shape of any given kidney protein, whether it be normal or mutated. 

鈥淲e call this AI model the digital twin of certain organs and their protein structures,鈥� Scalzo says. 鈥淭his AI modeling cuts down on trial and error. Instead of adjusting medications for side effects, in a few hours you determine that you want to target a particular gene or protein. AI will provide the structure of the compound and accordingly, the drug that you should make, which you would then test. That's the kind of pipeline we're building.鈥� 

AI- based modeling drug design from Nature Reviews Nephorology

Concerning drug design, any medication needs to bind to a specific cavity in a protein, grooves and gaps shaped in a way that could hold a small molecule, such as how a key fits into a lock. With AI, scientists can better determine the unique shape of the protein, or 鈥渢he shape of the lock,鈥� which makes designing drugs, 鈥渢he keys,鈥�  much quicker.

In addition, scientists now have better insight into the problems they鈥檙e targeting before designing the drug. Also important, AI modeling limits the reliance upon animal testing and years of drug trial and error with human beings. 

鈥淎I is not just chatbots taking over the world. In this case, scientists found a problem where AI could solve, going from sequence to structure,鈥� Wu adds. 鈥淚 think in these cases, it is extremely impactful to leverage AI to potentially save human lives.鈥� 

A Human Touch Still Needed 

Wu and Scalzo are careful to highlight the limitations of these AI models: they typically generate a single static snapshot of a protein rather than capture the full range of shapes it takes on inside a living cell, and they might generate overconfidence in researchers about proteins that are unusual or poorly represented in existing databases. 

The authors stress that AI predictions work best as a starting hypothesis. Results must be validated against real experimental data, particularly from techniques that can capture proteins in their natural cellular environment.

This research builds upon 每日大赛入口鈥檚 commitment to leading thought on ethical stewardship of AI. To guide the responsible use of AI for human good, for the past four years the University has hosted the on its Malibu campus. This year鈥檚 upcoming conference will be held from September 25 to 26 alongside the inaugural Global Summit on Faith in AI. 

鈥淎I can be used tangibly in a good way in many use cases,鈥� Wu concludes. 鈥淎s we get closer to AGI [artifical general intelligence], I'd rather see us focus on use cases like disease, where we'll eventually see a straight line from the AI model to a human getting better.鈥�

Learn more about 每日大赛入口鈥檚 vision for human-centered AI. View the published research in .