TPepPro: a deep learning model for predicting peptide-protein interactions.

Area of Science:

• Computational biology
• Drug discovery
• Bioinformatics

Background:

• Peptides show therapeutic promise, but studying peptide-protein interactions (PepPIs) is challenging.
• Experimental methods for PepPIs are costly and inefficient due to peptide flexibility.
• Existing computational methods for PepPI prediction require significant resources and lack accuracy.

Purpose of the Study:

• To develop an accurate and efficient computational model for predicting peptide-protein interactions (PepPIs).
• To address the limitations of current experimental and computational approaches in PepPI prediction.
• To facilitate the identification of potential peptide-based therapeutic agents.

Main Methods:

• Proposed TPepPro, a Transformer-based model for peptide-protein interaction (PepPI) prediction.
• Trained TPepPro on 19,187 peptide-protein complexes using sequential and structural features.
• Employed a combined local and global feature extraction strategy with an optimized neural network architecture (BN-ReLU).

Main Results:

• TPepPro achieved a prediction accuracy of 0.855, an 8.1% improvement over the next best model.
• TPepPro obtained an AUC of 0.922, significantly outperforming the second-best model's AUC of 0.844.
• The model successfully identified validated peptide-protein interactions, demonstrating its efficiency in detecting high-potential candidates.

Conclusions:

• TPepPro offers a computationally efficient and accurate solution for peptide-protein interaction prediction.
• The model aids in identifying promising peptide drug candidates for therapeutic applications.
• TPepPro's open-source availability facilitates further research and development in peptide drug discovery.