[Research on prediction model of protein thermostability integrating graph embedding and network topology features]

Area of Science:

• Computational biology
• Structural bioinformatics
• Machine learning in biochemistry

Context:

• Protein structure dictates function, making thermostability prediction crucial for applications.
• Current methods for predicting protein thermostability have limitations.
• Understanding protein stability is key to protein engineering and drug design.

Purpose:

• To develop a novel, high-precision method for predicting protein thermostability.
• To integrate graph embedding and network topological features for enhanced prediction accuracy.
• To explore the contribution of different features and algorithms in protein thermostability prediction.

Summary:

• A new method combines residue interaction networks (RINs), graph embedding (DeepWalk, Node2vec, Doc2vec), and network topological features using deep neural networks (DNN) and attention mechanisms.
• The model achieved 87.85% prediction accuracy on a bacterial protein dataset.
• Analysis revealed that combining DeepWalk, Doc2vec, and topological features is vital for identifying thermostable proteins.

Impact:

• Provides a practical and effective tool for protein thermostability prediction.
• Offers theoretical guidance for discovering new thermostable proteins and modifying existing ones.
• Facilitates advancements in protein engineering, enzyme design, and biotechnology.