Question pertaining to Normalization for Multiome Integration

[sylestiel]

@timoast

When INTEGRATING Multiome (RNA +ATAC) datasets do we need to normalize the RNA data and the ATAC data for each of the samples individually and after integration? Kindly clarify.

Thanks!

[timoast]

You can normalize gene expression data before integration and run RunTFIDF() after integration

[shahrozeabbas]

Hi, just wanted to clarify this further. The way I've been doing it is to normalize RNA and ATAC per dataset, identify top features for both modalities, merge, run pca/lsi, then harmony to remove batch effects.

Wouldn't you need to normalize ATAC with RunTFIDF() to integrate datasets? Here, I assume integration refers to correcting for batch effects.

[timoast]

If you're integrating the ATAC assay then you would need an LSI dimension reduction, and so would need to run TF-IDF beforehand. For multiome you could use the RNA assay to integrate the datasets, depends what you're aiming to do. However TF-IDF should be re-run after adding or removing cells to the dataset since that will change the IDF values for each peak.

[shahrozeabbas]

Okay got it, thanks!