Deep-ProBind: binding protein prediction with transformer-based deep learning model

Area of Science:

• Computational Biology
• Bioinformatics
• Molecular Biology

Background:

• Binding proteins are vital for cellular processes, regulating DNA, RNA, and peptide interactions.
• Identifying protein-binding peptides experimentally is costly and time-consuming.
• Existing sequence-based methods lack accuracy due to limited feature integration.

Purpose of the Study:

• To develop Deep-ProBind, a novel computational model for predicting protein-binding peptides.
• To integrate sequence and structural information for enhanced prediction accuracy.
• To provide a reliable tool for accelerating drug discovery and pharmacological research.

Main Methods:

• Utilized Bidirectional Encoder Representations from Transformers (BERT) and Pseudo position specific scoring matrix -Discrete Wavelet Transform (PsePSSM -DWT) for peptide encoding.
• Employed a transformer and evolutionary-based attention mechanism for feature extraction.
• Applied the SHapley Additive exPlanations (SHAP) algorithm for optimal feature selection and a Deep Neural Network (DNN) for classification.

Main Results:

• Deep-ProBind achieved 92.67% accuracy with tenfold cross-validation and 93.62% accuracy on independent samples.
• Outperformed existing models by 3.57% on training data and 1.52% on independent tests.
• Demonstrated high reliability and effectiveness in classifying protein-binding peptides.

Conclusions:

• Deep-ProBind offers a significant advancement in predicting protein-binding peptides.
• The integration of sequence and structural data enhances predictive performance.
• The model serves as a valuable resource for pharmacological studies and therapeutic development.