Simpson’s paradox confounds the interpretation of population-average protein and mRNA measurements. (a) For a particular gene, its protein levels across tissues can be poorly predicted by its mRNA levels, whereas the average protein levels can be well predicted by scaled mRNA levels.19 Thus mRNAs levels are unreliable surrogates for relative protein levels, and we need direct measurements of proteins. (b) Related manifestation of Simpson’s paradox indicates that the average levels of the ith and the jth proteins may appear positively correlated, even though they are inversely correlated within a cell type. Averaging across cells, even cell types sorted based on markers, will obscure the relationship between the ith and the jth proteins.