An Interpretable Machine-Learning Algorithm to Predict Disordered Protein Phase Separation Based on Biophysical...

Area of Science:

• Biochemistry
• Molecular Biology
• Biophysics

Background:

• Protein phase separation is crucial for biological organization and biomaterial formation.
• Intrinsically disordered protein regions (IDRs) are key drivers of this process.
• Existing prediction tools often lack interpretability of biophysical interactions.

Purpose of the Study:

• Introduce LLPhyScore, a novel predictor for IDR-driven protein phase separation.
• Provide a tool that interprets underlying biophysical interactions.
• Facilitate research into protein function and biomolecular condensates.

Main Methods:

• Developed LLPhyScore using sequence-based statistics from the RCSB PDB database.
• Trained the predictor on curated phase-separation-driving proteins and diverse negative sets.
• Analyzed contributions of various physical-chemical interactions.

Main Results:

• LLPhyScore effectively predicts protein phase separation with high recall.
• Identified key contributing features: solvent contacts, disorder, hydrogen bonds, pi-pi contacts, and kinked beta-structures.
• Electrostatics, cation-pi contacts, and lack of helical structure also play a role.

Conclusions:

• LLPhyScore offers a valuable resource for experimental guidance and hypothesis generation.
• Enables understanding of protein function in normal and pathological states.
• Aids in elucidating the emergence of specificity in biomolecular condensates.