Fig. 2.

Performance of different QC options on generation of PEER factors and sensitivity test for eGene detection using PEER and PCA. A The mean pair-wise correlation among the first 50 PEER factors. Each color and shape represent a specific cell type. B Time to generate 50 PEER factors by different quality control options on the pseudo-bulk matrix. C The x-axis denotes the number of PFs/PCs fitted as covariates in the association model. The y-axis represents the number of eGenes with at least one eQTL at local FDR < 0.05. The shape of each scatter point indicates whether generating PFs/PCs using all genes or the top 2000 highly variable genes (selected based on variance-to-mean ratio before transformation and scaling) to generate PEER factors (both excluded genes with ${\pi }_0\ge 0.9$, log(x + 1) transformed and standardized). The local regression lines are fitted for number of eGenes for number of latent variables. Different line types indicate four different scenarios