A prediction of mutations in infectious viruses using artificial intelligence

Area of Science:

• Virology
• Genomics
• Computational Biology

Background:

• Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) has evolved numerous subtypes with varying infectivity and severity.
• Regional and racial differences in SARS-CoV-2 mutations highlight the need for predictive models.

Purpose of the Study:

• To predict mutations during SARS-CoV-2 evolution.
• To identify key characteristics for accurate mutation prediction.
• To develop artificial intelligence models for forecasting novel viral mutations.

Main Methods:

• Collected and processed SARS-CoV-2 lineage, date, clade, and mutation data from public databases.
• Utilized artificial intelligence models, including XGBoost, with training sets based on clade information.
• Evaluated model performance with and without clade differentiation, and analyzed impacts of receptor-binding motif (RBM) mutations.

Main Results:

• Machine learning models incorporating clade differentiation achieved high predictive performance (XGBoost accuracy: 0.999).
• Reliance solely on mutation information yielded low model performance.
• Omicron's RBM mutations decreased predictive accuracy, but models successfully predicted potential mutations for clade 24C, including Q493.

Conclusions:

• Effective artificial intelligence models and predictive characteristics were developed for forecasting mutations in evolving viruses.
• Clade differentiation is a critical factor for accurate SARS-CoV-2 mutation prediction.
• The study provides a framework for anticipating future viral evolution and potential impacts on infectivity.