Viral Infection Detector: Ensemble Learning for Predicting Viral Infection in Single-cell Transcriptomics of Virus-Induced Cancers

Abstract

Oncoviruses cause around 15-20% of all human cancers. As 3 end-based single-cell RNA-sequencing (scRNA-seq) methods become more popular, the relationship between viral infection and tumor progression at the transcriptomic level has been increasingly studied at single-cell resolution. However, the identification of infected cells is challenging as viral reads from infected cells may not be captured, leading to high false negative rates. To recover these infected cells that have been missed, we developed VID, a stacked ensemble machine-learning model that predicts virus-infected cells in scRNA-seq datasets. Using VID, we uncovered biologically meaningful differences between infected and uninfected cells across different oncovirus-mediated cancers and different cell types. Our VID tool provides a user-friendly way to study changes induced by viral infections in scRNA-seq datasets.