AVPpred

AVPpred predicts antiviral peptides (AVPs) from peptide sequences to identify candidates with potential activity against human viruses such as influenza, HIV, HCV, and SARS.

Key Features:

• Extensive Database: Contains a curated collection of 1245 experimentally validated antiviral peptides active against human viruses including influenza, HIV, HCV, and SARS.
• Data Curation: Redundant peptides were removed to yield 1056 unique peptides, partitioned into a training set of 951 peptides and a validation set of 105 peptides.
• Feature Extraction: Extracts peptide sequence motifs, sequence alignment features, amino acid composition, and physicochemical properties for model input.
• Machine Learning Model: Employs a Support Vector Machine (SVM) trained using 5-fold cross-validation to develop predictive models.
• Performance Metrics: A model based on physicochemical properties achieved 85% accuracy and Matthew's Correlation Coefficient (MCC) of 0.70 in cross-validation, with validation accuracy of 86% and MCC of 0.71, outperforming general antimicrobial peptide prediction methods.

Scientific Applications:

• Peptide prioritization: Prioritizes AVP candidates for experimental validation and development of peptide-based antiviral therapies.
• Method comparison: Enables comparison against general antimicrobial peptide prediction methods to select AVP-specific leads.
• Outbreak response: Supports rapid identification of promising antiviral peptide candidates during viral outbreaks.

Methodology:

Feature extraction using peptide sequence motifs, sequence alignment, amino acid composition, and physicochemical properties; model development with a Support Vector Machine (SVM) evaluated by 5-fold cross-validation using a dataset of 1056 unique peptides split into 951 training and 105 validation peptides.