DP-site: A dual deep learning-based method for protein-peptide interaction site prediction.

Area of Science:

• Biochemistry
• Computational Biology
• Bioinformatics

Background:

• Protein-peptide interactions are crucial for understanding biological processes, drug discovery, and disease mechanisms.
• Experimental methods for identifying these interactions are often laborious, time-consuming, and costly.
• Predicting protein-peptide interactions computationally addresses these limitations.

Purpose of the Study:

• To develop an accurate and efficient computational framework for predicting protein-peptide interactions.
• To identify protein-peptide interactions at the residue level using diverse protein information.
• To overcome the drawbacks of experimental prediction methods.

Main Methods:

• The DP-Site framework employs a dual pipeline architecture with a combination predictor.
• Pipeline 1 uses a deep convolutional neural network for feature extraction and classification.
• Pipeline 2 integrates a deep long-short-term memory network and a random forest classifier, utilizing evolutionary, structure-based, sequence-based, and physicochemical features.

Main Results:

• DP-Site demonstrated robust and consistent performance on both ten-fold cross-validation and independent test sets.
• The method accurately predicts peptide binding residues in proteins.
• DP-Site significantly outperformed state-of-the-art sequence-based and structure-based methods, achieving a sensitivity of 0.770 and specificity of 0.799.

Conclusions:

• The DP-Site framework is proficient in predicting protein-peptide interactions.
• DP-Site surpasses existing methods in accuracy and efficiency.
• The DP-Site tool is publicly available for research use.