Machine learning techniques for sequence-based prediction of viral-host interactions between SARS-CoV-2 and human...

Area of Science:

• Computational biology
• Bioinformatics
• Machine learning in virology

Background:

• COVID-19, caused by SARS-CoV-2, is a global pandemic with significant mortality.
• Protein-protein interactions (PPIs) are crucial for SARS-CoV-2 infection mechanisms.
• Identifying virus-human protein interactions is key to understanding infection and developing treatments.

Purpose of the Study:

• To develop and validate machine learning models for predicting SARS-CoV-2 human protein interactions.
• To identify novel human protein targets for therapeutic intervention against COVID-19.
• To explore repurposable drugs for targeting predicted interactions.

Main Methods:

• Utilized sequence-based features of human proteins (amino acid composition, pseudo amino acid composition, conjoint triad).
• Developed various classification models, including an ensemble voting classifier (SVM-Radial, SVM-Polynomial, Random Forest).
• Validated predictions using biological experiments, gene ontology, and KEGG pathway enrichment analysis.

Main Results:

• The ensemble voting classifier demonstrated superior accuracy, precision, specificity, recall, and F1 score.
• Successfully predicted 1326 potential human target proteins for SARS-CoV-2.
• Identified several repurposable drugs targeting the predicted protein-protein interactions.

Conclusions:

• The study provides a robust method for identifying critical SARS-CoV-2 host targets.
• Predicted targets and drug candidates can accelerate the development of effective anti-COVID-19 therapies.
• This research facilitates further investigation into host-pathogen interactions for pandemic preparedness.