Protein Language Pragmatic Analysis and Progressive Transfer Learning for Profiling Peptide-Protein Interactions.

Area of Science:

• Computational Biology
• Bioinformatics
• Artificial Intelligence in Drug Discovery

Background:

• Protein complex structural data is rapidly expanding, presenting challenges for understanding protein function.
• Existing deep learning models often neglect crucial contextual information in protein sequences.

Purpose of the Study:

• To introduce interpretable interaction deep learning (IIDL)-peptide-protein interaction (PepPI), a novel deep learning model for peptide-protein interaction (PepPI) profiling.
• To address the limitations of current models in capturing complex contextual information within peptide and protein sequences.

Main Methods:

• IIDL-PepPI utilizes bidirectional attention modules to capture contextual information in peptides and proteins for pragmatic analysis.
• A progressive transfer learning framework is employed for simultaneous prediction of PepPIs and identification of binding residues.
• The model's performance is validated against state-of-the-art methods for predicting binary interactions and identifying binding residues.

Main Results:

• IIDL-PepPI demonstrates robust performance in accurately predicting peptide-protein binary interactions.
• The model effectively identifies critical binding residues involved in specific peptide-protein interactions.
• The model shows promise in peptide virtual drug screening and binding affinity assessment.

Conclusions:

• IIDL-PepPI offers a powerful, interpretable deep learning solution for in-depth peptide-protein interaction profiling.
• The model's capabilities are expected to significantly advance artificial intelligence-based peptide drug discovery.
• This approach is poised to enhance the elucidation of protein functions through detailed interaction analysis.