Musicality in protein interaction dynamics informs the multi-scale evolution of prosocial behaviour.

Like animal vocalization and display, human singing and dancing allows non-verbal establishment of behavioural co-relation (i.e. correlation) between individuals. The predictable mathematical structure of music is its most defining acoustic property, allowing human synchronization of both physical behaviour and emotion. In the biomolecular world, some proteins also interact in groups to achieve strong spatio-temporal co-relationships. This is prominent in amyloids, where many disordered fibrils individually conform to overall solenoid structures. We hypothesize that the vibrational frequencies captured during amyloid protein interactions may also exhibit elements of musicality related to this form of prosocial behaviour. Here, we develop a non-abstract data sonification method for computer-simulated molecular dynamic interactions. We apply auto-correlational and spectral cross-correlational analyses to a collection of sounds, defining 11 acoustic features that allow accurate machine learning classification of music from other types of natural sounds. By analysing statistical shifts in these correlative features defining musicality, we demonstrate that amyloid interactions are more speech-like and musical than less structurally conforming protein interactions, primarily due to significant shifts in memory (persistence) and first-order autocorrelation. We also find that music has less feature shift away from animal vocalization than human speech, suggesting it may have pre-dated the evolution of language.