Dimensionality reduction of 10x visium spatial datasets

scvi is an deep generative modeling for single-cell transcriptomics and can be used for dimensionality reduction (like PCA, its latent space in scvi) of scRNAseq. Now spatail transcriptomics is emerging and seurat/scanpy also apply PCA dimensionality reduction for spatial datasets (such as 10x visium), Can scVI’s latent space be used for that purpose too ?

We found scVI latent space to be useful in analysis of Visium experiments. The model assumptions of scVI apply to spot based sequencing technologies similarly (negative binomial, library size, generative process), depending on the data set quality we see high spot swaping in Visium and this might limit usability (SpotClean adjusts for spot swapping in spatial transcriptomics data | Nature Communications). This also applies to PCA.

1 Like