Computationally Detecting Viral Infection and Characterizing Host-Virus Dynamics in scRNA-seq Datasets

Abstract

Viruses pose significant threats to human health, with their impacts varying by type. Advances in single-cell RNA sequencing (scRNA-seq) have enhanced our understanding of viruses and host responses by mapping human and viral transcripts within individual cells. However, ambient RNA contamination complicates the accurate identification of viral infections in scRNA-seq datasets. To address this, we introduced scVirusFinder, a method that uses a zero-inflated negative binomial model followed by a support vector machine classifier to identify virus-infected cells. This approach improves the detection of true viral infections in scRNA-seq datasets of virus infected cells. We applied this method to scRNA-seq data from nasal washes of healthy donors and those with acute influenza during the 2017-18 season. We identified seventeen cell populations, including a novel epithelial cell population with high MHC class II gene expression in infected individuals. Influenza virus infections were found in most cell populations, primarily in epithelial cells and major immune cells such as macrophages and neutrophils. Using viral reads from the scRNA-seq data, we discovered that each donor harbored a unique influenza variant with distinct non-synonymous mutations. Additionally, we observed interferon production and response in infected samples, with type III interferon particularly produced in infected ciliated epithelial cells. This study highlights the challenge of identifying infected cells from scRNA-seq datasets and provides a robust solution applicable to clinical samples, enhancing our understanding of viral infections and paving the way for therapeutic discoveries.