Supplementary Figure 3: Other influences on measured NMD efficiency in human genomic data.

(a) Somatic mutations with high allelic frequency cause NMD estimates to appear more efficient. (b–e) Long exons and very large distances to the normal stop codon are associated with inefficient NMD in additional data sets consisting of TCGA somatic frameshift mutations (b,d) and Geuvadis germline truncating variants (c,e). (f,g) The standard EJC model and start-proximal NMD evasion have a dominant effect on NMD efficiency (f) and were thus factored out (g) using regression, facilitating discovery of further rules. (h,i) Rapid mRNA turnover attenuates the effects of NMD in somatic frameshifts (h) and germline variants (i). (j) NMD is slightly more efficient in highly expressed genes. (k) Reduced NMD efficiency in transcripts with short half-lives is also observed when using mRNA half-life measures from HeLa cells. The x axis shows the median mRNA half-life in minutes in each box plot. (l,m) The observed reduction in NMD efficiency in genes with fast mRNA turnover in B cells (l) is still observed upon explicitly factoring out gene expression levels from the NMD efficiency measure (m). In all panels with a continuous x axis, the axis was square root transformed. In all panels, the blue line is a fit using loess or generalized additive models (Online Methods) and the shaded area is its 95% confidence interval.