DeepAVPred: A Deep Learning Framework Integrating RECM-DWT and BiGRU for Accurate Antiviral Peptide Prediction

Viral infections continue to pose a significant global health challenge. Antiviral peptides (AVPs) have emerged as promising therapeutic agents owing to their high specificity, favorable safety profiles, and broad-spectrum antiviral activity. However, experimental AVP identification is time-consuming and costly, while existing computational predictors often have limitations in predictive accuracy and feature representation. Here, we propose DeepAVPred, a deep learning-based framework for AVP identification. DeepAVPred integrates complementary sequence descriptors to capture physicochemical properties and pairwise residue interactions. We present a novel feature representation, the Residue Energy Content Matrix augmented with a Discrete Wavelet Transform (RECM-DWT). RECM-DWT characterizes multi-scale patterns in residue-energy information and complements conventional sequence descriptors by providing wavelet-derived features across multiple resolution levels. The integrated features are then modeled using an attention-based Bidirectional Gated Recurrent Unit (BiGRU) network to learn contextual dependencies and informative sequence patterns. On the AVPtrain dataset, DeepAVPred achieved strong performance in 10-fold cross-validation, with an accuracy of 0.990, a Matthews Correlation Coefficient (MCC) of 0.980, and an area under the receiver operating characteristic curve (AUC) of 0.993. On the independent test dataset, AVPtest, DeepAVPred achieved an accuracy of 0.942, an MCC of 0.886, and an AUC of 0.949. These results demonstrate the predictive performance of DeepAVPred and support its potential as a computational framework for prioritizing candidate antiviral peptides for further experimental validation.