Fig. 5

PEER factors are almost identical to PCs in GTEx eQTL and sQTL data [10]. a The y-axis shows all 49 tissue types with GTEx QTL analyses ordered by sample size (from small to large). Given a fully processed molecular phenotype matrix, we summarize the correlation matrix (in absolute value) between the PEER factors obtained and used by GTEx and the top PCs into two numbers: the average of the diagonal entries and the average of the off-diagonal entries. With the exception of Kidney - Cortex sQTL data, the diagonal entries have averages close to one, and the off-diagonal entries have averages close to zero (both have minimal standard errors). b A typical correlation heatmap showing near-perfect one-to-one correspondence between the PEER factors and the top PCs. c In Kidney - Cortex sQTL data, the PEER factors and the top PCs do not have a perfect one-to-one correspondence. The reason is because the PEER factors are highly correlated with each other (d), while PCs are always uncorrelated (Additional file 1: Section S5.1). The numbers in parentheses represent sample sizes. To produce this figure, we reorder the PEER factors based on the PCs (Additional file 1: Algorithm S1), although in almost all cases, this reordering does not change the original ordering of the PEER factors because PEER initializes with PCs [24]