DeepAntiVP: Deep Learning Framework For Predicting Antiviral Peptides Using Transformed Features

Viruses represent a pervasive threat to global public health. Despite substantial advancements in antiviral research, the rapid development of novel vaccines and therapeutic interventions remains a critical challenge. Antiviral peptides (AVPs) have emerged as promising candidates within the field of immunoinformatics, offering new directions for drug discovery. However, the mechanistic understanding of their broad-spectrum biological activities is still incomplete. The post-genomic era has yielded extensive peptide sequence data, facilitating the development of computational approaches for automated AVP identification. Nevertheless, the predictive accuracy of existing machine learning-based AVP classifiers remains limited, primarily due to suboptimal feature engineering and less effective learning algorithms. To address these limitations, we introduce DeepAntiVP, a deep learningbased framework for precise AVP identification from peptide sequences. The model integrates hybrid feature representations encompassing physicochemical properties, compositional attributes, and energy-based metrics. These features are derived using advanced extraction techniques, including component protein sequence representation (CPSR), extended pseudo amino acid composition (ExPseAAC), Residue Energy Context Matrix-based (RECM). The resulting feature vectors are utilized by a deep learning architectures, notably a deep forest model and a deep bidirectional gated recurrent unit (Deep_BiGRU) classifier. Among these, the Deep_BiGRU consistently demonstrated superior performance across multiple benchmark datasets, outperforming existing state-of-the-art AVP prediction methodologies. The success rates of the proposed AVP-based medthod on independent test are: Acc=0.941, Sn=0.987, Sp=0.900, AUC=0.949 and MCC=0.886, respectively outperforming the existing advanced AVP models in the literature.