Proteins can also populate other states in addition to their native states. At the high concentrations present in the cell most proteins form condensed states, which can be solid-like, known as the amyloid state and liquid-like, referred to as the droplet state. Thus, we should consider the droplet state as a fundamental state of proteins along with the native state and the amyloid state. Rapidly accumulating evidence indicates that a wide range of proteins, perhaps even most of them are capable to form the droplet state through a liquid-liquid phase separation (LLPS) process.

Condensed states of proteins, both the liquid-like membraneless organelles and solid-like aggregates, contribute in fundamental ways to the organisation and function of the cell. While cellular activities associated with the native state rely on the individual properties and specific interactions of the polypeptide chains, a wide range of functions associated with the droplet and amyloid states emerge from collective behaviours of these molecules. Therefore, under cellular conditions it necessary generalizing the well-established structure–function paradigm towards a wider state–function paradigm to account for the vast repertoire of functionalities enabled by collective, generic interactions.

Overall, high conformational entropy and multiple binding configurations are characteristic of the droplet state. A wide variety of sequence motifs can drive the formation of liquid-like condensates, which is primarily determined by its sequence complexity. These motifs contribute to non-native, disordered interactions, such as different combinations of charged, dipole and quadrupole interactions, exemplified by cation-π or hydrophobic contacts.

Formation of the droplet state is driven by disordered interactions of protein regions, which sample many binding configurations. Such disordered binding modes can be mediated by both structured and disordered protein regions lacking a local sequence bias, which stabilises the folded state. Disordered binding modes can be achieved by redundant, multivalent motifs, as well as without a distinguished recognition sequence. It is the sequence complexity, which determines the binding properties, whether a protein region folds or remains disordered in the bound form.

Prediction of liquid-liquid phase separation is based on predicting the probability of disordered binding modes. The FuzDrop approach can identify droplet-promoting regions, which can sample disordered interactions in droplets. Based on the number and propensity of such regions, and accounting for non-specific hydrophobic forces, the probability of undergoing liquid-liquid phase separation can be estimated. Droplet-drivers are proteins, which can spontaneously undergo liquid-liquid phase separation, whereas droplet-clients possess droplet-promoting regions, which require an additional partner or specific conditions to induce liquid-liquid phase separation.