LIBD DLPFC Development and Aging Browser

We modeled differential expression across age at each of the five feature summarizations (gene, exon, junction, transcript, and ER) in the 320 control subjects across the lifespan. We modeled expression, after transforming with log2 with an offset of 1, as a function of age after creating using linear splines with breakpoints at ages: birth (0), 1, 10, 20, and 50, further adjusting for sex and ancestry/ethnicity (first 3 MDS components). F-statistics were computed comparing the model containing age (including the linear splines), sex, and ethnicity, to a statistical model with just sex and ethnicity, with corresponding p-values calculated based on an F-distribution with 7 and 308 degrees of freedom, and Bonferroni adjustment within each feature type was performed using the number of features with non-zero expression (gene RPKM > 0.01, exon RPKM > 0.1, and junction RP80M > 0.2 with non-novel annotation) across all samples as the number of tests (which varied by feature type). We also computed post-hoc statistics on the data, including the Pearson correlation between “cleaned” expression (after regressing out the effects of sex and ethnicity, holding the age effects constant), and age to determine if the expression of the fetal rose or fell across the lifespan, and also measured the fetal versus postnatal log2 fold changes. Additional details are available at Jaffe et al, bioRxiv 2017.