Multi-AOP: a lightweight multi-view deep learning framework for antioxidant peptide discovery.

Area of Science:

• Biochemistry and Bioinformatics
• Computational Chemistry
• Artificial Intelligence in Drug Discovery

Background:

• Antioxidant peptides (AOPs) show potential for disease prevention and food preservation due to their free radical scavenging properties.
• Traditional experimental methods for discovering AOPs are inefficient and resource-intensive.

Purpose of the Study:

• To develop an efficient computational framework for enhanced antioxidant peptide discovery.
• To integrate sequence and structural information for improved AOP prediction accuracy.

Main Methods:

• Developed Multi-AOP, a lightweight multi-view deep learning framework utilizing Extended Long Short-Term Memory (xLSTM) for sequence embeddings.
• Employed Message Passing Neural Network (MPNN) on SMILES representations to extract molecular graph features, capturing physicochemical properties.
• Implemented hierarchical fusion of sequence and graph features for comprehensive peptide analysis.

Main Results:

• Multi-AOP achieved high prediction accuracies: 0.8043 (AnOxPePred), 0.9684 (AnOxPP), and 0.9043 (AOPP).
• The framework consistently outperformed conventional machine learning and state-of-the-art deep learning approaches.
• A unified AOP dataset was created to foster the development of generalizable AOP prediction models.

Conclusions:

• The Multi-AOP framework offers a significant advancement in efficient and accurate antioxidant peptide discovery.
• Integrated sequence and graph learning provides a powerful approach for predicting peptide functionalities.
• Publicly accessible datasets and models will accelerate future research in AOP development.