A sequence-based model for identifying proteins undergoing liquid-liquid phase separation/forming fibril aggregates...

Area of Science:

• Biochemistry
• Molecular Biology
• Computational Biology

Background:

• Liquid-liquid phase separation (LLPS) and amyloid fibril formation are critical protein aggregation processes linked to health and disease.
• Understanding the sequence-level differences between LLPS and amyloidogenic proteins is crucial but underexplored.

Purpose of the Study:

• To systematically investigate and compare sequence-level features distinguishing proteins undergoing LLPS from those forming amyloid fibrils.
• To develop a predictive model for classifying proteins based on their aggregation behavior (LLPS, amyloid fibrils, or background).

Main Methods:

• Comparative analysis of 36 sequence-derived features between LLPS and amyloid fibril proteins.
• Development of random forest-based classification models (binary and three-class).
• Feature selection and ablation analysis to identify key predictive features.

Main Results:

• Significant differences were found in 24 out of 36 sequence features between LLPS and amyloid fibril proteins.
• The fraction of intrinsically disordered residues (F_IDR) was the most critical feature in binary classification.
• Cysteine and leucine composition were significant in three-class classification (LLPS-Fibrils-Background).
• A six-feature model (FLFB) achieved an average AUC of 0.83 for predicting protein aggregation states.

Conclusions:

• Sequence features can effectively differentiate between proteins undergoing LLPS and those forming amyloid fibrils.
• Machine learning models, particularly FLFB, offer a powerful tool for predicting protein aggregation pathways from sequence data.
• This research provides insights into the molecular basis of distinct protein aggregation mechanisms.