Thresholding CITE-Seq-based Antibody Expression Data to Improve Cell Type Identification in scRNA-Seq

Abstract

Abstract scRNA-seq is widely used to assess immune cell heterogeneity by measuring individual cell transcriptomes. Surface markers have been used for decades to type immune cells by flow and more recently mass cytometry. Expression of surface markers is poorly correlated with the mRNA levels of their encoding transcripts. CITE-Seq provides a systematic approach to analyze surface phenotypes along with cell transcriptomes by using oligonucleotide-tagged antibodies. Like flow cytometry, CITE-seq-based assessment of surface markers is complicated by non-specific binding. Here, we introduce and validate a systematic method of thresholding the levels of protein marker expression to address the issue of non-specific antibody binding. To remove the non-specific signal, a threshold value separating noise from expression for each surface marker is obtained by comparing the antibody signal in a cell type known to not express this surface marker. If no negative cell type is available, we use deconvolution of overlapping normal distributions to find the best separator in density plots (Ridgeline plots) by deconvolution (normalmixEM tool) of overlapping normal distributions. In total, 34 of 49 antibodies tested were reliably thresholded, thus effectively removing the noise from the signal. In a validation experiment testing 188 antibodies at 4 different concentrations, 124 were successfully thresholded, including all 34 mentioned above. Supported by the National Institutes of Health (NIH) P01 HL136275