ITP-Pred: an interpretable method for predicting, therapeutic peptides with fused features low-dimension...

Area of Science:

• Biotechnology and Pharmaceutical Sciences
• Computational Biology
• Bioinformatics

Background:

• The peptide therapeutics market presents significant opportunities for the biotechnology and pharmaceutical sectors.
• Accurate identification and property exploration of therapeutic peptides are crucial for drug development.
• Existing machine learning models for therapeutic peptide prediction often lack detailed explanations of their decision-making processes.

Purpose of the Study:

• To develop an interpretable machine learning model for predicting therapeutic peptides.
• To enhance the understanding of factors influencing therapeutic peptide recognition.
• To provide guidance for designing improved therapeutic peptide prediction models.

Main Methods:

• Developed an Interpretable Therapeutic Peptide Prediction (ITP-Pred) model.
• Engineered three feature descriptors: amino acid composition (AAC), group AAC, and coding autocorrelation.
• Integrated features and employed a Convolutional Neural Network-Bi-directional Long Short-Term Memory (CNN-BiLSTM) architecture.

Main Results:

• ITP-Pred demonstrated superior prediction performance compared to existing methods on benchmark datasets.
• Cross-validation and independent verification confirmed the model's high accuracy.
• Analysis revealed key sequence order and physicochemical properties important for prediction.

Conclusions:

• The ITP-Pred model offers a robust and interpretable approach to therapeutic peptide identification.
• The model's insights into feature importance can guide the rational design of novel therapeutic peptides.
• This work complements existing methods by providing a transparent and effective tool for peptide drug discovery.