Tetramer protein complex interface residue pairs prediction with LSTM combined with graph representations

Area of Science:

• Computational Biology
• Structural Biology
• Bioinformatics

Background:

• Protein-protein interactions are crucial for biological processes.
• Predicting interaction sites is key to understanding protein function.
• Few methods exist for predicting interfaces in protein multimers.

Purpose of the Study:

• To develop a deep learning method for predicting interface residue pairs in protein tetramers.
• To address the challenges of predicting interfaces in complex protein multimers.

Main Methods:

• Developed a deep network combining Long Short-Term Memory (LSTM) and graph neural networks.
• Utilized protein structure data from the Protein Data Bank.
• Extracted physicochemical and geometric features of surface residues.
• Transformed protein multimer data into topological graphs for analysis.

Main Results:

• Successfully predicted protein tetramer interaction interface residue pairs.
• Demonstrated the validity of the method using various evaluation indicators.
• The graph-based LSTM approach effectively handles Non-Euclidean structure data.

Conclusions:

• The developed deep learning model accurately predicts interface residue pairs in protein tetramers.
• This method offers a valuable tool for studying complex protein assemblies.
• Advances the understanding of protein-protein interactions in biological systems.