Cytokine activity estimation and receptor abundance approximation with multimodal scRNA-seq/CITE-seq data.

Single-cell RNA-sequencing (scRNA-seq) data enables individual cell resolution quantification of messenger RNA. While revolutionary for revealing key cell type and cell phenotype-specific heterogeneity, scRNA-seq data has important statistical challenges. First, scRNA-seq data is usually more sparse and second, it is more variable than bulk transcriptomics data. Given these challenges, more intuitive and interpretive statistical and computational methods are needed to develop appropriate solutions. Here, we detail the development of three techniques to perform receptor abundance estimation and cytokine activity estimation for scRNA-seq data. While SPECK (Surface Protein abundance Estimation using CKmeans-based clustered thresholding) and STREAK (gene Set Testing-based Receptor abundance Estimation using Adjusted distances and cKmeans thresholding) methods address the sparsity constraints of scRNA-seq data via dimensionality reduction and a thresholding mechanism and co-expression analysis using joint scRNA-seq/CITE-seq data, respectively, SCAPE (Single cell transcriptomics-level Cytokine Activity Prediction and Estimation) aims to leverage a cytokine signaling activity database via a modified gene set testing approach to accommodate negative weights. Our approaches work well in practice and aim to provide more interpretive solutions to statistical challenges of scRNA-seq data. Furthermore, our methods have the potential to be integrated in bioinformatics pipelines for the tasks of cell type and cell state identification.