Proteins bearing prion-like domains (PrLD) are widespread throughout all kingdoms of life. These domains resemble the intrinsically disordered, low complexity, Q/N-rich regions present in most yeast prions. Multiple studies have predicted around 1% of the human proteome corresponds to these prion-like proteins. Characterization of this human protein subset has stated its enrichment in DNA and RNA binding proteins and their involvement in the formation of biomolecular condensates. These transient membraneless compartments phase separate through highly dynamic liquid-liquid demixing and are related to several neurodegenerative diseases. This is particularly evident in cases of naturally occurring mutations that increase the aggregation propensity of PrLDs by converting these liquid compartments into solid aggregates, compromising their dynamic nature. Hence there is a need for in silico tools able to quantify the impact of mutations on the aggregation propensities of this kind of disease-associated proteins.
The debate of whether the self-assembling propensities of prion-like proteins depend only on a biased amino acid composition accounting for the whole PrLDs, or instead on specific sequential features facilitating their transition to amyloid-like states is an unsolved hot topic in our field. Nonetheless, we have recently shown that a function that takes into consideration both parameters predicts better the impact of a wide spectrum of punctual and multiple mutations or deletions on the aggregation of the model ALS-associated prion-like hnRNPA2 protein. Accordingly, we introduce the AMYCO (combined AMYloid and COmposition based prediction of prion-like aggregation propensity) webserver which implements this approach to allow fast and automated predictions.