A bunch of keys lies on the table: each opens only one lock. In living cells, nanobodies often play the role of such keys: tiny antibody fragments that stick to strictly defined proteins. They can be used to deliver a drug exactly to a diseased cell or to block a harmful process. But shaping the "bit"—the three flexible loops responsible for recognition — used to be slow and imprecise. Either variants were tested for days, or it was impossible to target a specific region of the protein.

The authors of the new study created the program EasyNano. It solves this task in 10–20 minutes on a powerful home computer. Instead of blindly testing millions of combinations, the program smoothly "files down" the shape of the loops, step by step bringing them closer to the target. EasyNano relies on a ready-made folded model of the target protein and tracks how the distances between amino acids change, like fitting a key to a lock's cast. The calculations use a lightweight version of the ESMFold2 model (721 million parameters)—a fast structure predictor. The full version (1.3 billion parameters) ensures that the nanobody's scaffold does not distort.

An important finding was where to start the fitting. If you start from the natural sequence of the loops rather than a random set, the program finds better solutions. An experienced locksmith doesn't carve a key from scratch but refines an existing blank — the same principle applies here. Tests on six different nanobody–target pairs showed that the improvements are not random. For one example, statistical significance reached 5.7 sigma — far beyond ordinary fluctuations. Moreover, several runs with different starting conditions lead to different but equally good loop shapes. This means nature has several ways to build a reliable key.