Accounting for this redundancy resulted in a total of 7107 transcript/TE pairs, which make up our dataset for analyzing the effect of TE insertions of expression levels.
Several non-parametric statistical tests [27], [28] have been recently recommended for comparing different classifiers on multiple datasets (accounting for the effects of multiple comparisons) [29].
For each causal get set, 1000 simulation datasets were generated for the effect size 0 to compute type I error and 100 simulation datasets for effect size 0.3 and 0.6 to compute powers.
Kruskal-Wallis tests confirmed this effect of orientation (p = 0.0267) for the entire dataset, but the effect is insignificant after the removal of tandem duplicates (p = 0.2894).
For the combined dataset, the effect was statistically significant (p < 0.001; Fig. 1b), as for the Belgian data alone [ 2].
For the full dataset, the effect was only significant in the first year of follow-up which may reflect increased risk earlier in treatment or insufficient power to detect an association after the first year.
For microarray datasets examining the effect of injecting miR-430 into MZDicer embryos (Giraldez et al., 2006 ) or knocking out miR-155 in T cells (Loeb et al., 2012 ), we required the mean signal intensity of an mRNA to exceed 3.0 and 2.5, respectively.
PCA analysis using this dataset normalized for the organ effect resolved two temporal phases of the transcriptomic response to cytokinin.
Adjusting the 29K dataset for batch effects introduced by the manufacturing procedure using ComBat as described above significantly improved the sample correlations (median = 0.62, P = 2.9 e − 11), indicating that the overall gene expression patterns had become more alike.
For Illumina BeadArray datasets examining the effect of either knocking down miR-302/367 (Lipchina et al., 2011 ) or transfecting miR-522 (Tan et al., 2014 ), we required the mean signal intensity to exceed 7.5 and 7.0, respectively.
