An ensemble-based model comprising deep learning for predicting peptide-binding residues in proteins.

Area of Science:

• Computational Biology
• Bioinformatics
• Drug Discovery

Background:

• Protein-peptide interactions are crucial for cellular functions and implicated in diseases like cancer.
• Experimental methods for studying these interactions are costly and time-consuming.
• Existing computational methods often lack sufficient predictive accuracy.

Purpose of the Study:

• To develop a highly accurate computational model for predicting protein-peptide interactions.
• To integrate structural and sequence-based features for improved prediction.
• To provide a valuable tool for functional genomics and drug discovery.

Main Methods:

• Developed PepENS, an ensemble model combining deep learning and traditional machine learning.
• Integrated features such as half-sphere exposure and position-specific scoring matrices.
• Utilized embeddings from a pre-trained protein language model for sequence analysis.

Main Results:

• PepENS achieved high performance on test datasets, with AUC scores of 0.860 (Dataset 1) and 0.846 (Dataset 2).
• Demonstrated superior precision and AUC compared to state-of-the-art methods.
• Achieved precision of 0.596 on Dataset 1 and 0.539 on Dataset 2.

Conclusions:

• PepENS offers a significant advancement in predicting protein-peptide interactions.
• The model's accuracy supports its utility in functional genomics and accelerating drug discovery.
• The PepENS software and datasets are publicly available for research use.