A Review of CITE-Seq Best Practices in the CAR-T Landscape

Abstract

The purpose of this study was to determine a set of best practices for analyzing CITE-seq data, particularly in the context of CAR-T therapies. To determine the best method of denoising protein expression data, multiple single-cell cell surface protein processing pipelines, including a custom pipeline, were run on datasets from three different input samples: CAR-T final product, PBMCs and BMMCs. Of the methods tested, scAR, a machine-learning tool for the denoising of ambient protein expression, was determined to be the best pipeline for CITE-seq processing. After denoising the protein expression, the effect of mutations based on mRNA variant detection on protein expression was investigated using cb_sniffer, a tool designed for mutation calling from single cell data with low read depth. The mutations that were observed in these data did not appear to have a significant effect on the cell surface protein expression.