| To hula hoop, you stand and rotate your hips to the right and to the left. To ride a bike, you crouch, outstretch your arms and pedal your legs. To dive into a pool, you extend your arms, tuck your chin and lean forward. Those body shapes enable us to take certain actions — or, as a biologist might say, our structure determines our function. This is true, too, at the molecular level. Every imaginable task performed by a cell has a protein that's built to execute it. By some estimates, there are 20,000 different types of proteins in the human body: Some blood cell proteins are perfectly shaped to grasp oxygen and iron molecules, some skin cell proteins provide structural support, and so on. Each one has a shape fitted to its job. However, if a protein misfolds, it can no longer perform, and that can lead to dysfunction and disease. By figuring out how proteins fold, biologists would not only gain a deeper understanding of the proteins themselves but also potentially unlock new ways to target disease-related proteins with new medicines. It has turned out to be a formidable scientific challenge. Every protein starts off as a string of smaller linked molecules called amino acids. As the amino acids line up in the order dictated by a gene, they bend and buckle into the protein's proper shape within microseconds — a phenomenon that astounded 20th-century scientists when they discovered it. In the 1950s, the biochemist Christian Anfinsen hypothesized that there must be an internal code built into the string of amino acids that directs how a protein should fold. If that was the case, he thought, there should be a way to predict a protein's final structure from its amino acid sequence. Making that prediction became known as the protein folding problem. Since then, some scientists have redefined the problem as three related questions: What is the folding code? What is the folding mechanism? Can you predict the protein structure just by looking at its amino acid sequence? Biologists have spent decades trying to answer these questions. They have experimented with individual proteins to understand their structures and built computer programs to deduce any patterns in how proteins fold. They've studied the physics and chemistry of amino acid molecules down to the atomic level to glean the rules of protein folding. Even so, biologists have made only halting progress on understanding a protein's internal folding rules since Anfinsen stated the problem. A few years ago, they had a breakthrough, when new artificial intelligence tools opened up part of the problem. The tools, most prominently Google DeepMind's AlphaFold, can't explain how a protein folds from a string of amino acids. But given an amino acid sequence, they can often predict the final shape it folds into. Only in the coming decades will it become clear whether that distinction — knowing how a protein folds versus what it folds into — will make a difference in applications like drug development. Does the magician need to reveal the magic trick?
What's New and Noteworthy Early in May, Google DeepMind announced the latest iteration of their protein-predicting algorithm, called AlphaFold3, which predicts structures not only for single proteins but also for proteins bound to each other and for other biomolecules like DNA and RNA. As I reported for Quanta, this announcement came just months after a competing protein prediction algorithm — RosettaFold All-Atom, which was developed by the biochemist David Baker at the University of Washington School of Medicine and his team — announced a similar upgrade. "Now you're getting at all the complex interactions that matter in biology," Brenda Rubenstein, an associate professor of chemistry and physics at Brown University, told me. Still, there's a long way to go before these algorithms can determine the dynamic structures of proteins as they move within cells. Sometimes proteins act unpredictably, throwing another wrinkle into the folding problem. Most proteins fold into a single stable shape. But as Quanta reported in 2021, some proteins can refold into multiple shapes to perform multiple functions. These fold-switching proteins are not well studied, and no one knows how abundant they are. But with advances in technologies like cryo-electron microscopy and solid-state nuclear magnetic resonance, researchers are seeing them more clearly. What's more, some proteins have regions that don't fold into a discrete shape but wiggle dynamically. As Quanta reported in February, these intrinsically disordered proteins can have important functions, like enhancing the activity of enzymes, the class of proteins that bring about chemical reactions. When proteins misfold, they can cluster together and wreak havoc on the body. Protein aggregates are hallmarks of neurodegenerative diseases like Alzheimer's, in which possibly toxic protein aggregates called amyloid plaques collect between neurons and undermine the brain's signaling. As Quanta reported in 2022, protein aggregation may be widespread in aging cells; understanding why proteins misfold and pile up could help with the development of treatments for aging-related issues. Sometimes misfolded proteins can also promote the misfolding and aggregation of other proteins, setting off a cascade of detrimental effects that exemplifies just how critical it is for a protein to bend into its proper shape. | 
