The difference between merge and integration with Seurat objects

Question

[re_raz]

Hi everyone I have two questions: When can we use merge for Seurat objects? When can we use integration for Seurat objects? I have two datasets from different experiments. each dataset has cancer samples and healthy samples. To do clustering, I did the following: 1- I created a Seurat object for each sample, then I calculated QC and filtered the cell. 2- normalized each object with SCT 3- combined samples from each experiment together using merge() and then I made a list of two objects (two experiments) 4- applied reference-based integration to combine the two datasets. I used two samples (cancer and health) with high cell number from each dataset as a reference dataset. 4- Run PCA, UMAP, Find Neighbour, FindCluster.

This workflow worked, however, I am unsure if what I have done correctly in biology?

For one dataset with many samples, Could I apply the Seurat integration workflow instead of merge()?

Answer 1

[rpolicastro]

You should only use merge for technical replicates, and in theory for a group of samples with a low batch effect. Integration in Seurat (and related) was developed because there tends to be a relatively strong batch in the manifolds. By this I mean that even if two cell populations are the same between two samples, they will appear as two partially or fully separate clusters. Integration tries to "smooth out" the differences in batches so that cells that are likely similar will cluster together. As with anything single-cell related I would suggest exploring and validating the results from both merging and integration to get a better feel for your data.

[re_raz]

Thank you for your explanation. I would like to add this point, the two datasets are from different platforms. one experiment used the Seq-Well platform, whereas another used the 10× Genomics platform. Will not applying integration be better here?