AttnSeq-PPI: Enhancing protein-protein interaction network prediction using transfer learning-driven hybrid...

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Protein-protein interaction (PPI) networks are crucial for cellular functions.
• Experimental and traditional computational methods for PPI study have significant limitations.

Purpose of the Study:

• To develop an accurate and efficient sequence-based deep learning framework for predicting protein-protein interactions.
• To overcome the limitations of experimental and structure-based computational methods.

Main Methods:

• Proposed AttnSeq-PPI, a deep learning framework utilizing a hybrid attention mechanism.
• Employed ProtT5 language model for protein sequence embedding.
• Combined self-attention and cross-attention to capture inter-protein features and long-range dependencies.

Main Results:

• Achieved 99% accuracy on human and multi-species datasets.
• Demonstrated superior generalization and performance compared to existing models.
• Successfully predicted novel PPIs with high precision and reduced false negatives.

Conclusions:

• AttnSeq-PPI offers a powerful, sequence-based approach for PPI prediction.
• The framework provides a valuable tool for understanding cellular processes and identifying potential drug targets.
• A web-based tool is available for accessible PPI prediction.