iACVP

iACVP predicts anticoronavirus peptides (ACVPs) from peptide sequences to identify candidate antiviral peptides for coronaviruses.

Key Features:

• Input dataset: Uses a dataset comprising antiviral peptides (AVPs) and experimentally verified anticoronavirus peptides (ACVPs).
• Feature types: Combines conventional sequence-derived features with word-embedding models such as word2vec (W2V).
• Embedding dictionary: Employs a dataset-specific W2V dictionary generated from training and independent test datasets rather than a general proteome database like UniProt.
• k-mer optimization: Conducts a systematic search to identify the optimal k-mer value for W2V embeddings.
• Algorithms evaluated: Systematically evaluates Transformer, Convolutional Neural Network (CNN), bidirectional Long Short-Term Memory (BiLSTM), Random Forest (RF), and Support Vector Machine (SVM) models.
• Best-performing model: Identifies Random Forest (RF) combined with W2V as consistently outperforming other classifiers across datasets.
• Discrimination objective: Enhances discrimination between positive (ACVPs) and negative peptide samples.
• Application goal: Predicts putative ACVPs to prioritize candidates for experimental validation and reduce experimental screening effort.

Scientific Applications:

• Antiviral candidate discovery: Prioritizes peptide sequences as putative ACVPs to guide discovery of antiviral agents against coronaviruses.
• Experimental prioritization: Reduces time and cost by triaging candidates for experimental validation.
• Method benchmarking: Provides a comparative framework for evaluating machine learning approaches in peptide antiviral prediction.

Methodology:

Uses a dataset of AVPs and experimentally verified ACVPs, extracts conventional sequence features and word2vec (W2V) embeddings with a dataset-specific W2V dictionary and optimized k-mer, and evaluates Transformer, CNN, BiLSTM, RF, and SVM models, finding RF combined with W2V to perform best.