Peptide Resources

DeepAVP is a dual-channel deep neural network ensemble for antiviral peptide (AVPs) prediction, designed to analyze variable-length peptide sequences. The innovative architecture includes: ① an LSTM channel capturing long-range sequence dependencies for efficient variable-length data processing; ② a CONV channel building dynamic neural networks to analyze local evolutionary features. The model fine-tunes amino acid substitution matrices for functional peptides to enhance feature representation.

Deep-AVPpred is a deep learning-based webserver for antiviral peptide (AVPs) prediction, leveraging transfer learning to address the high cost and inefficiency of natural AVP extraction. The model uses pre-trained neural networks to extract sequence features, fine-tuned for virus infection (e.g., COVID-19)-related peptides, achieving 94% validation accuracy and 93% test accuracy—outperforming state-of-the-art models. Key advantages include: ① enhanced generalization in small-sample scenarios via transfer learning; ② successful identification of novel AVPs from human interferon-α family proteins, providing drug candidates; ③ visual interface to display sequence feature weights for mechanism interpretation. Supporting batch peptide sequence upload, the server enables efficient virtual screening for human and veterinary antiviral compounds.

DeepAVP-TPPred is an efficient machine learning model for antiviral peptide (AVPs) identification, addressing limitations of existing methods in feature engineering, accuracy, and generalization. The model employs a two-step strategy: ① generating two novel transformed feature sets via self-designed image-based extraction algorithms, integrated with evolutionary information features (e.g., position-specific scoring matrix); ② optimizing feature space using Binary Tree Growth Algorithm to select discriminative subsets for deep neural network classification.

DeepB3P3 is a novel framework for blood-brain barrier penetrating peptides prediction, achieving high-precision prediction via deep learning model.

DeepCPPred is a two-layer deep learning framework using elastic net for feature optimization and cascade deep forest for high-accuracy identification of cell-penetrating peptides and their uptake efficiencies.