High throughput data such as RNAseq is often complicated by technical errors that are not explicitly modeled in the experimental design and can confound the detection of differential expression. Differences in counts from samples run on different days or different locations or on different machines are an example of a technical error that is very often present and which we should try to model in our experimental design. An approach to this is to build a surrogate variable into our experimental design that explains the batch effect and take it into account in the modeling and differential expression analysis stages. We'll use the SVA package to do this.