Overview
The batch alignment and analysis workflow builds on the ‘simple discovery’ workflow by adding a step to facilitate batch alignment. This workflow allows for the correction of technical variation or shifts in signal levels in samples stained and/or acquired across multiple batches. To do this, we have implemented the CytoNorm algorithm (Van Gassen 2020). CytoNorm uses reference control samples that are prepared and recorded along with each batch of samples to identify and correct technical variations between individual batches, while preserving biologically relevant differences. For more information on CytoNorm, see Van Gassen et al 2020, and for more information on our implementation in Spectre, see Ashhurst et al 2021.
The demo dataset used for this worked example are cells extracted from mock- or virally-infected mouse bone marrow, measured by flow cytometry. Expression level values in these datasets have been manipulated to simulate acquisition over two batches.
Reference controls for CytoNorm
An example of reference control samples are aliquots of peripheral blood mononuclear cells (PBMCs) that are derived from a single donor at one time point, and cryopreserved (i.e. multiple aliquots of a biologically identical sample). Each time a set of PBMC samples from the study cohort are thawed, stained, and recorded, a reference controls is also thawed, stained, and recorded. Differences in signal level between the reference controls allows CytoNorm to learn the differences in signal levels due to the batches, and correct them, while preserving biological differences between the individual samples. In our demo dataset, we are using bone marrow samples derived from separate mice. Though not derived from the same mouse, these are similar enough that they can be used successfully as reference controls.