A protein in a cell is not a rigid part but rather a soft glove that can grasp different objects. Until now, protein designers worked like sculptors, crafting a key for a single lock: one target, one shape. But life is more complex: sometimes a protein needs to recognize two different viruses at once or change its contours depending on its neighbors.
The new method, Chamaileon, tackles this challenge. Instead of drawing a protein from scratch, it learns from examples where the amino acid sequence (beads on a string), the three-dimensional fold, and the context, what the protein should bind to or what state it should be in, are all known. Training is like the work of a translator: the model sees a context–structure pair and reconstructs the sequence text, then works in reverse. In this way, it captures the hidden rules by which the same chain can fold differently.
At the application stage, the model samples many variants of a single sequence, fitting it to different contexts, like a tailor adjusting one suit to different body shapes by slightly tweaking the pattern. This technique, called Mixture-of-Paths Sampling, helps overcome the scarcity of experimental data where all possible forms of a protein are captured.
The authors tested Chamaileon on a specially assembled benchmark, CROSS, ranging from binding two different proteins to switching between open and closed forms. In all cases, the generated sequences were stable and recognized the intended targets. This approach paves the way for designing smart proteins that don't just hit one spot but flexibly respond to their environment — almost like living things.
Nature already has transformer proteins: for instance, calmodulin changes shape when it binds calcium and regulates dozens of processes in the cell.
In Robert Ibatullin's novel 'The Rose and the Worm', a biotechnology is described that creates proteins with switchable functions to adapt to the environment — much like Chamaileon, but on the scale of a whole organism.