Integrating SCTransform-normalized data with LIGER

[minigel]

Hello, I am wondering whether Seurat objects that have been normalized with SCTransform, rather than log-normalization, can be integrated with LIGER. The current vignette only describes the latter workflow: https://htmlpreview.github.io/?https://github.com/satijalab/seurat.wrappers/blob/master/docs/liger.html

SCTransform outputs Pearson residuals that are used for integration with Seurat: (https://satijalab.org/seurat/v3.1/integration.html#sctransform). Can these residuals also be used as input to LIGER?

[skpalan]

Technically, yes. You can follow that tutorial starting from RunOptimizeALS as long as you have all the required data stored in the corresponding slots such as scale.data.

On the other hand, in order to perform liger correctly on your data, you need to manually make sure the Pearson residuals (which are stored in the scale.data slot) are non-negative.

Also the tutorial and wrapper functions provided by seurat-wrapper are pretty old and not compatible with the newest version of liger, we are currently connecting with Satija Lab to update both the package and tutorials. Please pay attention to future updates that are going to occur very soon.

[minigel]

Thank you so much for your response! I look forward to the updates to the tutorial and wrapper functions.

In the meantime, would you be able to suggest manual transformations of the Pearson residuals that would be compatible with NMF/LIGER -- e.g. would adding a constant to all the residuals be appropriate?