To tell one protein from another, you need to describe its shape and composition. There are dozens of such descriptions: molecular length, charge, solubility. But not all are equally useful. A method called Differentiable Information Imbalance (DII) can pick out the most informative features while discarding the rest.

The authors of the study treated the selected features as coordinates of an imaginary physical system. It turned out that as the number of features changes, the system undergoes a phase transition, like water freezing into ice. For physicochemical features, the transition is sharp: from a “glassy” state with a rigid structure to a “liquid” state with more freedom. For structural features, the transition is more blurred.

The most interesting part: the critical number of features found by DII coincides with the point after which protein classification accuracy stops improving. In other words, physics tells you when you have enough data. This offers a simple and objective way to avoid overloading models with unnecessary measurements and to save computational resources.

So glass physics helps bring order to protein description, and to pick exactly the features needed for reliable recognition.