Multi-ACPNet: A multi-scale sequence-structure feature fusion framework for anticancer peptide identification and...

Area of Science:

• Computational Biology
• Bioinformatics
• Drug Discovery

Background:

• Anticancer peptides (ACPs) show high efficacy and low drug resistance for cancer therapy.
• Current ACP identification methods focus on sequence data, neglecting crucial structural information.
• Simultaneous prediction of ACP identification and functional activity remains a challenge.

Purpose of the Study:

• To develop a novel dual-function predictor, Multi-ACPNet, for both ACP identification and activity classification.
• To integrate both sequence and spatial structural features for enhanced prediction accuracy.
• To overcome the limitations of existing ACP prediction methods.

Main Methods:

• A multi-stage framework integrating sequence and structural features.
• Hybrid Bidirectional Long Short-Term Memory (BiLSTM) and causal convolutional networks for sequence pattern analysis.
• Multi-scale Graph Convolutional Network (GCN) for dynamic fusion of structural dependencies.

Main Results:

• Multi-ACPNet achieved high accuracy (0.8140-0.9536) for ACP identification across three datasets.
• Functional prediction yielded strong performance with AUC 0.9033 and F1-score 0.8472.
• The model significantly outperformed existing state-of-the-art predictors.

Conclusions:

• Multi-ACPNet effectively integrates sequence and structural data for accurate ACP identification and functional prediction.
• This dual-function approach offers a promising advancement in cancer therapy development.
• The model provides a robust tool for discovering and characterizing novel anticancer peptides.