A HEK293T-derived explainable CatBoost signature for estimating HCoV-OC43 viral burden from host transcriptomes

Area of Science:

• Computational Biology
• Virology
• Genomics

Background:

• Human coronavirus OC43 (HCoV-OC43) is a model betacoronavirus, but limited for SARS-CoV-2 studies.
• Understanding host responses to viral infections is crucial for developing therapeutic strategies.

Purpose of the Study:

• To develop a predictive model for viral burden using host gene expression.
• To identify key host genes associated with viral replication intensity.
• To differentiate viral infection states from general cellular stress.

Main Methods:

• Utilized CatBoost regression on single-cell RNA sequencing data (12,980 HEK293T cells).
• Employed SHAP values for interpreting gene impact on predictions.
• Validated the model on independent datasets, including bulk RNA-seq and non-viral stress conditions.

Main Results:

• Achieved high predictive performance (R²=0.870) for viral burden at single-cell resolution.
• Identified 10 core host genes, with TPI1 as the top predictor of infection intensity.
• Demonstrated model's ability to distinguish high-burden viral infection from tunicamycin-induced cellular stress.

Conclusions:

• The developed pipeline systematically identifies interpretable transcriptional signatures of viral replication.
• The model accurately predicts viral burden and differentiates viral states from general stress.
• This approach aids in understanding host-virus interactions and developing targeted interventions.