Identification of Phase Separating Proteins With Distributed Reduced Alphabet Representations of Sequences

Area of Science:

• Biochemistry
• Computational Biology
• Molecular Biology

Background:

• Protein phase separation is crucial for cellular functions like bacterial division and tumorigenesis.
• Understanding the molecular drivers of phase separation, such as hydrophobicity and protein dynamics, is key.
• Identifying new phase separating proteins can advance cellular physiology knowledge and disease intervention.

Purpose of the Study:

• To develop a sequence-based machine learning model for predicting phase separating proteins.
• To leverage existing data on phase separating proteins for predictive modeling.

Main Methods:

• Utilized reduced alphabet schemes based on hydrophobicity and conformational similarity.
• Employed distributed representations of protein sequences and biochemical properties as input features.
• Applied Support Vector Machine (SVM) and Random Forest (RF) algorithms on curated and balanced datasets.

Main Results:

• The Random Forest model, trained on a balanced dataset incorporating hydropathy, conformational similarity, and biochemical properties, achieved 97% accuracy.
• Highlighted the importance of conformational similarity (amino acid flexibility) and hydrophobicity in prediction.

Conclusions:

• Developed an accurate, sequence-based machine learning method for predicting phase separating proteins.
• Demonstrated the utility of interpretable features like conformational similarity and hydrophobicity.
• Suggests that incorporating these features can further enhance prediction performance for phase separation.