amyloid-predict and LLPS-predict: Predicting phase separation propensities in the intrinsically disordered proteome

Area of Science:

• Cellular Biology
• Biophysics
• Computational Biology

Background:

• Amyloid formation and liquid-liquid phase separation (LLPS) are crucial cellular processes implicated in both normal functions and diseases.
• Understanding the propensity for these phenomena is vital for biological and medical research.

Purpose of the Study:

• To introduce a computational framework for predicting amyloid and LLPS propensities using protein language models.
• To enable rapid, proteome-wide annotation of peptides and residues for these properties.
• To investigate the distribution and biological relevance of amyloid and LLPS propensities in the human proteome.

Main Methods:

• Development of computational classifiers (amyloid-predict and LLPS-predict) based on protein language model embeddings.
• Benchmarking against existing AI and physics-based tools for classification performance.
• Application of classifiers to intrinsically disordered regions (IDRs) across the entire human proteome.

Main Results:

• The amyloid-predict tool demonstrated superior classification performance and speed compared to existing methods on a hexapeptide benchmark.
• Analysis revealed specific protein categories enriched in amyloid and/or LLPS propensities, including signaling receptors, carbohydrate-binding proteins, and mRNA-binding proteins.
• Identified patterns of both high amyloid and LLPS propensity in certain amyloid-forming and prionic proteins.

Conclusions:

• The developed framework provides a rapid and accurate method for assessing amyloid and LLPS potentials across proteomes.
• The study offers novel insights into the biological roles of protein aggregation and LLPS by mapping their distribution in disordered proteins.
• This work serves as a valuable tool for basic research, disease mechanism studies, and the development of peptide therapeutics.