MIPPIS: protein-protein interaction site prediction network with multi-information fusion

Area of Science:

• Biochemistry
• Computational Biology
• Bioinformatics

Background:

• Protein-protein interaction site prediction is vital for understanding biological processes and drug development.
• Existing methods face challenges including long processing times, high costs, and low accuracy.

Purpose of the Study:

• To develop a novel, accurate, and efficient network for predicting protein-protein interaction sites.
• To address the limitations of current prediction methodologies.

Main Methods:

• A multi-information fusion network is proposed, integrating features from position-specific scoring matrix, hidden Markov model, dictionary of protein secondary structure, and one-hot encoding.
• A multi-channel approach extracts deep amino acid features using graph convolutional networks (GCN) for spatial information, bidirectional long short-term memory (BiLSTM) for primary structure, and ProtT5 for comprehensive embeddings.

Main Results:

• The proposed network effectively extracts spatial and sequential features from proteins.
• Multi-channel feature extraction provides robust amino acid representations.

Conclusions:

• The novel network demonstrates superior performance over existing structure-based and sequence-based methods across key evaluation metrics.
• The findings highlight the model's effectiveness in protein-protein interaction site prediction.