Single-Cell Gene Expression Modeling

Raw data were analyzed using Fluidigm Real-Time PCR analysis software. Preprocessing and data analysis were conducted using KNIME 2.11.2 and RStudio Version 0.99.486. Preprocessing with a linear model to correct for confounding factors was conducted as previously described (39 (link)). To model the bimodal gene expression of single cells from T cell clones, the Hurdle model, a semi-continuous modeling framework, was applied to the preprocessed data (40 (link)). This allowed us to assess the differential expression profiles with respect to the frequency of expression and the positive expression mean via a likelihood ratio test. Genes were ranked by a score derived from the number of comparisons with statistically significant differences in gene expression and the number of donors with statistically significant comparisons (0.5 points per significant comparison multiplied by the number of donors showing significant differences in expression of each gene). Genes with the same score were ranked based on the delta of their median expression differences between the cognate and control stimuli.