Extended Data Fig. 3: Illustration of 14 real cell-state abundance traits used in our non-null simulated GWAS for T cells.

For each trait in Supplementary Fig. 9, we plot the true cell-level phenotype for which we simulated associated genotypes with varying levels of noise. Each UMAP includes one dot per T cell in the OneK1K dataset. Cells that do not affect the trait are colored grey. For example, for the “CD8 Naive Program 1” trait, we used a gene expression program that varies substantially across naive CD8 + T cells. We quantified the usage of that program across all naive CD8 + T cells and defined the trait value per individual in the dataset as the mean use of that gene expression program across all naive CD8 + T cells in that individual’s sample. Therefore, for the “CD8 Naive Program 1” trait we color all cells that are not naive CD8 + T cells grey because they do not influence the trait and we color each naive CD8 + T cell by its use of the gene expression program. Cells with greater use of the program are colored deeper red and cells with less use of the program are colored deeper blue.