Hi,

I am working on analyzing multiple scRNA-seq dataset from embryonic tissues at progressive stages. I used three recent integration algorithms 1) liger, 2) Seurat CCA and 3) fastMNN. I started with these based on recommendation from peers and availability of Seurat Wrapper for these approaches.

In my experience, I observe that both liger and Seurat CCA are over integrating dataset. For example, I expect to see a some overlapping population and some unique when I comapre a tissue sample at two different points; both of these approaches suggest that two populations are very identical.

Just merging the two data sort of gives the expected results. fastMNN approach also give similar result. I feel that some adjustments need to be made for integrating two data at different points, however current integrating approaches (liger and seurat CCA) overcorrect the problem. And I don't know what is a good approach.

I just want know what other people who are dealing with integrating multiple scRNA-seq think about it?

Thanks!

it's a pain of our generation, really. Everyone gets by as a function of the desired result, sometimes even forgoing the integration/batch-correction altogether and using the normalized logcounts directly if the samples were more or less prepped and sequenced together