iACP-DPNet: a dual-pooling causal dilated convolutional network for interpretable anticancer peptide identification

Area of Science:

• Computational biology
• Bioinformatics
• Drug discovery

Background:

• Anticancer peptides (ACPs) show promise for cancer therapy due to their specificity and safety.
• Current ACP discovery is hindered by costly experimental validation and limited computational models.
• Existing models face challenges with small datasets, poor interpretability, and weak generalization.

Purpose of the Study:

• To develop a robust and interpretable deep learning model for predicting anticancer peptides (ACPs).
• To address limitations of existing ACP prediction models, including dataset size and feature learning.
• To enhance the efficiency and accuracy of identifying potential ACPs for cancer treatment.

Main Methods:

• Constructed a larger, diverse dataset by consolidating existing ACP literature and databases.
• Developed the iACP-DPNet model using ProtBert for feature extraction and LightGBM/MIC for feature selection.
• Implemented a dual-pooling mechanism (GlobalAveragePooling and attention pooling) with causal dilated convolutions.
• Utilized t-SNE, ISM, and SHAP for model interpretability analysis.

Main Results:

• The iACP-DPNet model achieved high performance metrics: Sp 96.1%, Sn 92.91%, Acc 94.5%, MCC 89.05% on a novel dataset.
• Demonstrated superior performance compared to existing state-of-the-art methods in comparative analyses.
• Showcased strong generalizability by outperforming other models on an additional independent dataset.

Conclusions:

• iACP-DPNet provides an effective and interpretable framework for anticancer peptide (ACP) prediction.
• The model's advanced architecture and feature learning enhance prediction accuracy and efficiency.
• This research offers a valuable tool for accelerating the discovery of novel ACPs for cancer therapy.