Automated denoising of CITE-seq data with ThresholdR.

Abstract

Cellular indexing of transcriptomes and epitopes by sequencing (CITE-seq) is a potent addition to single-cell RNA sequencing (scRNA-seq). This method enriches transcriptomic insights by incorporating information about the cell surface phenotype through the application of oligonucleotide-tagged monoclonal antibodies. Similar to observations in flow cytometry, the CITE-seq signal (antibody-derived tag [ADT]) contains technical noise originating from ambient antibodies within the reaction compartment, non-specific binding, and/or imperfect titration. To denoise ADT data provided through CITE-seq experiments, we present ThresholdR, an R-based automated tool, to reliably and systematically find the threshold that separates the signal from the noise for each antibody. We assess the performance of ThresholdR across different datasets and platforms and benchmark it against two alternative methods, DSB (denoised and scaled by background) and CellBender. We show that ThresholdR remedies the high false negative rates of DSB and CellBender. We propose that denoising with ThresholdR can improve cell-type annotation and improve downstream analyses.