General scRNA-seq workflow using batches and the simulation strategy and evaluation in this paper. (A) The fundamental steps of the batch correction workflow. Data is ingested in samples, combined, and preprocessed in various ways, depending on the method. The data is then commonly projected onto a custom embedded space, possibly with lower dimension than the original data, where some correction takes place; correction can also be performed on the original count data or k-NN graph. The corrected object is then returned, depending on the method, a count matrix, lower dimensional embedding, NN graph, etc. Downstream methods then use this corrected object. (B) The general workflow of the evaluation conducted in this paper. scRNA-seq data is split randomly into two batches. Batch correction methods are applied to these two pseudobatches. The change in the data, at different stages of the scRNA-seq workflow pipeline, is then measured to assess the change that the act of batch correction has had on the data.