Structure-based machine-guided mapping of amyloid sequence space reveals uncharted sequence clusters with higher...

Area of Science:

• Biochemistry
• Computational Biology
• Structural Biology

Background:

• Amyloid conformation is adopted by diverse sequences, but the full extent of amyloid sequence space remains undefined.
• Current knowledge is biased towards hydrophobic, beta-sheet prone sequences found in globular protein cores and yeast prions.

Purpose of the Study:

• To explore amyloid sequence space beyond its established boundaries using a machine learning approach.
• To identify novel sequence types with amyloid-forming potential.

Main Methods:

• Utilized high-resolution structural data from the protein databank.
• Implemented a machine learning approach named Cordax.
• Applied t-Distributed Stochastic Neighbour Embedding (t-SNE) for clustering sequence data.

Main Results:

• The Cordax approach expanded the exploration of amyloid sequences beyond hydrophobic regions.
• Identified clusters with lower aliphatic content and higher charge, and regions with helical and disordered propensities.
• Demonstrated that amyloid propensity can be uncoupled from solubility.

Conclusions:

• Amyloid formation is not limited to hydrophobic, beta-sheet prone sequences.
• Discovered novel sequence types compatible with surface-exposed protein patches, functional amyloids, and liquid-liquid phase transitions.
• Cordax provides a new framework for exploring the broader landscape of amyloid sequences.