Machine learning methods accurately predict host specificity of coronaviruses based on spike sequences alone.

Area of Science:

• Virology
• Genomics
• Bioinformatics

Background:

• Coronaviruses, including MERS, SARS-CoV-1, and SARS-CoV-2, pose significant threats due to interspecies transmission.
• The spike (S) protein is critical for coronavirus host specificity, mediating cell receptor attachment.
• Accurate prediction of coronavirus host specificity is vital for early outbreak detection and control.

Purpose of the Study:

• To evaluate the utility of coronavirus spike protein sequences in predicting host specificity.
• To develop and assess machine learning models for host specificity prediction based on spike sequences.

Main Methods:

• Analysis of 1238 coronavirus spike protein sequences.
• Utilizing t-SNE embeddings to visualize sequence segregation based on host and virus species.
• Applying machine learning algorithms including Support Vector Machines (SVM), Logistic Regression, Decision Tree, and Random Forest.

Main Results:

• Spike sequences clustered effectively in t-SNE embeddings according to host and/or virus species.
• Machine learning models achieved high performance metrics (0.95-0.99 accuracy, F1 scores, sensitivity, specificity).
• Decision Tree analysis identified biologically significant protein regions associated with host specificity.

Conclusions:

• Coronavirus spike protein sequences alone are sufficient for predicting host specificity.
• Machine learning models offer a powerful tool for identifying potential zoonotic threats and understanding viral evolution.
• This approach can enhance strategies for managing and preventing future coronavirus outbreaks.