Severe batch effect after denoising.

[EddieLv]

Hi, I have tested SPAGATE-tensorflow on 2 replicates(2 slices) of one stage, but when I load the denoised count matrices into Seurat, I found severe batch effect than raw count matrices.
Can you give some advice on it?

[QIFEIDKN]

I think it might be because you trained STAGATE for each section individually. In this scenario, we recommend concatenating the spatial networks of different sections together and processing them with the same STAGATE model.

For you question, we added two additional tutorials to introduce how STAGATE work for multiple sections (with or without batch effects). See https://stagate.readthedocs.io/en/latest/AT1.html and https://stagate.readthedocs.io/en/latest/AT2.html for details.

By the way, recently, we extended our STAGATE to identify spatial domains on multiple sections, and designed a new algorithm STAligner, which may be online within a month. : )

[EddieLv]

Thanks for your reply~ And I am really looking forward to your new algorithm:)

[EddieLv]

Hi, now I've solved the batch effect through your new tutorial:)
But I have another question, can I load the denoised count matrix into Seurat and do normalize-scale-pca-clustering? Or should I cluster based on the embedding matrix (SATGATE), and only use the denoised count matrix to view gene-expression and DE-analysis, etc. ?

[QIFEIDKN]

In my opinion, the denoised results are only used to view gene expressions. Actually, like other denoising or imputation methods (like MAGIC or scVI), the return matrix of STAGATE is a dense matrix, which may lead the p-value of DE-analysis get lower, giving many false positives.

[EddieLv]

Got your point, thanks~