PMC:4157143 / 20590-23154
Annnotations
{"target":"https://pubannotation.org/docs/sourcedb/PMC/sourceid/4157143","sourcedb":"PMC","sourceid":"4157143","source_url":"https://www.ncbi.nlm.nih.gov/pmc/4157143","text":"We hypothesized that rare variants acting either alone or in combination with common variants can cause an eQTL to exhibit a larger effect size in the family than in the population. To identify such cases, we applied a ranking scheme in which we compared gene-expression cis-eQTLs between the family and the population to find genes that exhibited larger effect sizes within the family (see Material and Methods). At CI \u003e 0.95 (or empirical p value \u003c 0.05), we found that 319 (including both paternal and maternal β measurements) of the 7,341 genes we tested had effect sizes exceeding that of the best population cis-eQTL SNP (false-discovery rate [FDR] = 7,341 × 0.05 × 2 / 319 \u003e 1; Figure 1A). Using comparisons of β, we did not find more relatively large-effect eQTLs than we would expect by chance; however, we identified that this FDR is likely overconservative primarily because of differences in noise between the family and population (see Figures S17–S19), and we therefore also discuss less conservative estimates of FDR (see Figures S17–S19). It is important to note that FDR here measures whether there are more large effects in the family than in the population; however, ranking relative effect sizes by empirical p values is biologically meaningful whether there is an excess or a depletion. Such relative effects overlap (to a degree) genes measured only by absolute effect size in the family; for instance, when comparing genes at the 95% percentile for absolute β versus relative β, we observed an overlap of 52% (Figure S12). However, we chose to use in all subsequent analyses the ranking of genes according to their relative effect sizes instead of absolute effect sizes because we hypothesized that the former might better inform family-specific effects. By instead measuring fit (R2), we identified 577 cis-eQTLs that had a better fit in the family than the best population-level cis-eQTL variant (CI \u003e 0.95; FDR = 7,341 × 0.05 / 577 = 63%; Figure S10). Among those genes that exhibited the largest effect sizes and fits in the family (both at a CI \u003e 0.95), there was a significant overlap of 36.4% (Figure S11). To exclude the possibility of technical factors underlying effect-size differences, we repeated the analysis by using different quantification pipelines (Figures S13 and S14), population discovery-panel sizes (Table S6), and alternative methods for choosing the best SNP (Figure S16); we observed no significant difference in the discovery set of large-effect genes or on further downstream analyses (see Material and Methods).","tracks":[]}