A two-task predictor for discovering phase separation proteins and their undergoing mechanism

Area of Science:

• Cell Biology
• Biochemistry
• Computational Biology

Background:

• Liquid-liquid phase separation (LLPS) is a key cellular mechanism for organizing biomolecules into membraneless organelles.
• Identifying proteins that undergo LLPS is essential for understanding cellular compartmentalization and function.

Purpose of the Study:

• To develop a computational tool, Opt_PredLLPS, for identifying potential phase separation proteins.
• To predict the interaction-dependent mechanism of LLPS for identified proteins.

Main Methods:

• A two-task deep learning model combining CNN and BiLSTM for initial LLPS protein prediction.
• An XGBoost model with feature selection for predicting interaction-dependent LLPS mechanisms.
• In silico saturation mutagenesis to identify critical phase separation regions.

Main Results:

• Opt_PredLLPS effectively identifies potential LLPS proteins and differentiates their mechanisms.
• The study validated the model's performance on benchmark datasets.
• Key protein regions involved in phase separation were identified.

Conclusions:

• Opt_PredLLPS provides a valuable resource for discovering phase separation proteins.
• The findings advance the understanding of LLPS mechanisms and cellular organization.
• This tool can guide future research in biomolecular condensates.