Batch effects were then adjusted by the COMBAT method [ 30].
Before fit model (1), inverse normal transformation was applied to expression level to remove outlier's effect, and batch effects were removed by adjusting principal components calculated based on all genes expression (13, 27).
Because DNA samples were randomly allocated for methylation analysis, characteristics such as age, sex, and smoking categories were equally represented on each plate; consequently, although batch effects were statistically significant, adjusting for batch effects had very little impact on the associations between smoking behaviors and methylation intensity.
Expression values for each time point (adjusted for batch effects) were pulled for the model for the 139 probe sets matched to our leaf apoplast proteins of interest (see below).
Kaplan Meier survival analysis (univariate) and Cox proportional hazards modeling (adjusted for age, body mass index, and stratified for batch effects) were used to study association of tertiles of metabolites on all-cause mortality in TwinsUK.
Batch effects were removed by aligning the within-batch medians for all measurements.
The bottom row of Table 1 (which shows the numbers of differentially expressed genes between batches within a single experiment using ANOVA) shows that after batch effects have been adjusted for, there are no differentially expressed genes between batches for the disease samples, or between batches for the healthy control samples.
Batch effects between datasets were adjusted for by the empirical Bayes method ComBat/SVA [ 32– 32].
The ability of the method to correctly adjust for batch effects is shown in Figure 4. Fold change values are lower in the adjusted data set relative to the unadjusted data set for all but seven of the 600 expressed genes, again indicating that the method adjusts the fold changes in the correct direction.
The ability of the method to correctly adjust for batch effects is further illustrated in Figure 3.
