AVP-HNCL: Innovative Contrastive Learning with a Queue-Based Negative Sampling Strategy for Dual-Phase Antiviral Peptide Prediction.
Journal:
Journal of chemical information and modeling
Published Date:
Jun 6, 2025
Abstract
Viral infections have long been a core focus in the field of public health. Antiviral peptides (AVPs), due to their unique mechanisms of action and significant inhibitory effects against a wide range of viruses, exhibit tremendous potential in protecting organisms from various viral diseases. However, existing studies on antiviral peptide recognition often rely on feature selection. As data volume continues to grow and task complexity increases, traditional methods are increasingly showing limitations in feature extraction capabilities and model generalization performance. To tackle these challenges, we propose an innovative two-stage predictive framework that integrates the ESM2 model, data augmentation, feature fusion, and contrastive learning techniques. This framework enables simultaneous identification of AVPs and their subclasses. By introducing a novel top-k queue-based contrastive learning strategy, the framework significantly improves the model's accuracy in distinguishing challenging positive and negative samples and its generalization performance. This approach provides robust theoretical support and technical tools for advancing research on antiviral peptides. Model evaluation results show that on Set 1-nonAVP, the framework achieves an accuracy of 0.9362 and a Matthews correlation coefficient (MCC) score of 0.8730. On the Set 2-nonAMP, the model achieves perfect accuracy (1.0000) and an MCC score of 1.0000. In addition, during the second stage, the model accurately predicts the antiviral activity of antiviral peptides against six major virus families and eight specific viruses. To further enhance accessibility for users, we have developed a user-friendly web interface, available at http://www.bioai-lab.com/AVP-HNCL.
