Building on the successful applications of physical concepts and sampling algorithms introduced in the study of disordered systems, in particular artificial neural networks, we quantitatively explore how well a quantity known as the local entropy describes the native state of model proteins.

We found indications that real proteins have higher local entropy. If we follow the analogy with deep learning, where training would not be possible if it did not happen in a landscape with a pervasive cluster of solution, we can think that also evolution can not happen from random proteins that are not evolvable enough. Local entropy seems a good candidate to describe the evolvability of proteins even before evolution. On top of this, it may be that evolution also optimizes local entropy.