Data-driven discovery of antiviral peptides against PRRSV using multiple machine learning models.

Area of Science:

• Biochemistry and Molecular Biology
• Computational Biology and Bioinformatics
• Virology

Background:

• Protein and peptide network analysis is crucial for understanding biological processes and predicting antivirulence.
• Antiviral peptides (AVPs) show broad-spectrum anti-virulence capabilities, but current databases lack sufficient precise annotations.
• Porcine reproductive and respiratory syndrome virus (PRRSV) poses a significant threat, necessitating novel therapeutic strategies.

Purpose of the Study:

• To screen differentially expressed proteins and peptides in healthy versus PRRSV-infected porcine tissues.
• To predict novel AVPs using machine learning (ML) and deep learning (DL) computational methods.
• To identify potential AVP therapeutics against PRRSV.

Main Methods:

• Proteomics was employed to quantify proteins and peptides from lung, small intestine, and large intestine samples.
• Machine learning (ML) and deep learning (DL) models were developed using physicochemical features like amino acid composition, secondary structure, and hydrophilicity.
• A deep learning graph neural network (GNN) was utilized and benchmarked against conventional ML models, including random forest (RF) and support vector machine (SVM).

Main Results:

• Lysine, arginine, and leucine were identified as significant predictive features for AVPs.
• The random forest (RF) model achieved the highest Area Under the Curve (AUC) of 0.95 ± 2, outperforming GNN and SVM models (0.94 ± 1).
• Amino acid composition was found to be a key predictive factor in identifying AVPs.

Conclusions:

• Integrating proteomics with computational modeling successfully identified peptides with antiviral potential against PRRSV.
• The RF model demonstrated superior discriminative power for AVP prediction.
• These findings provide valuable resources for experimental validation and development of PRRSV AVPs as therapeutics.