PreAntiCoV

PreAntiCoV identifies anti-coronavirus (anti-CoV) peptides using machine learning to support discovery of peptide candidates against SARS-CoV-2.

Key Features:

• Dataset Utilization: Incorporates diverse negative datasets including antivirus peptides without anti-CoV function (antivirus), antimicrobial peptides lacking antivirus function (non-AVP), and peptides lacking antimicrobial properties (non-AMP).
• Machine Learning Approach: Employs random forest classifiers combined with imbalanced learning strategies to address severe class imbalance in training data.
• Performance Metrics: Reports geometric mean (GMean) of sensitivity and specificity of 83.07% versus antivirus peptides, 85.51% versus non-AVPs, and 98.82% versus non-AMPs.
• Two-Stage Classifier: Implements a double-stage classification with stage one distinguishing AMPs from regular peptides (AUCROC 97.31%) and stage two distinguishing anti-CoV peptides among AMPs (GMean 79.42% on independent test sets).

Scientific Applications:

• Therapeutic development: Facilitates identification of candidate anti-CoV peptides for development of novel coronavirus-directed therapeutics.
• Virology research: Supports research efforts aimed at inhibiting coronaviruses and studying peptide-based antiviral activity.

Methodology:

Uses random forest classifiers with imbalanced learning strategies on datasets including antivirus, non-AVP, and non-AMP negatives; applies a two-stage classification (AMP vs regular peptides, then anti-CoV vs other AMPs) and evaluates performance using AUCROC and GMean on independent test sets.