Package Compatibility

1. Introduction

This tutorial demonstrates how to convert Seurat or AnnData/h5ad object to SPATA2. SPATA2 provides the convenient function asSPATA2() to load objects from the mentioned packages.

2. Seurat

First, an exemplary 10X Visium dataset from the mouse brain is loaded using SeuratData. We are following the Seurat tutorial in the following.

# load packages for this tutorial 
library(Seurat)
library(SeuratData)
library(tidyverse)
library(SPATA2)
library(SPATAData)

# create an example Seurat object
# install and load data with SeuratData
InstallData("stxBrain")
brain <- LoadData("stxBrain", type = "anterior1")

# process
brain <- NormalizeData(brain, assay = "Spatial")
brain <- SCTransform(brain, assay = "Spatial")
brain <- RunPCA(brain, assay = "SCT", verbose = FALSE)
brain <- RunUMAP(brain, reduction = "pca", dims = 1:30)

brain <- FindNeighbors(brain, reduction = "pca", dims = 1:30)
brain <- FindClusters(brain, verbose = FALSE)

# left plot
SpatialPlot(brain, group.by = "seurat_clusters")

# right plot
DimPlot(brain, group.by = "seurat_clusters")

For proper conversion information about certain aspects are required.

2.1 The sample name

The sample_name argument uniquely identifies the object and dataset within a group of multiple SPATA2 objects.

2.2 The spatial method

The platform argument specifies the platform used, which is integral to several SPATA2 functions as it stores details about the underlying spatial method of the data. For a comprehensive understanding of what constitutes a platform or spatial method in SPATA2, refer to the documentation with ?spatial_methods. Although the default setting is platform = 'Undefined', it is recommended to specify this argument with a valid input. This specification is necessary because the platform used cannot always be accurately inferred from a Seurat object alone.

## [1] "MERFISH"     "SlideSeqV1"  "Undefined"   "VisiumSmall" "VisiumLarge"
## [6] "VisiumHD"    "Xenium"

2.3 Assay name and assay modality

The assay_name argument indicates which assay from the Seurat object should be used. In many cases, only one assay is present, making the default setting of assay_name = NULL sufficient. However, note that Seurat and SPATA2 handle assays slightly differently. For example, using the SCTransform() function in Seurat creates a new assay within the Seurat object. Despite these differences, both contain gene expression data.

names(brain@assays)
## [1] "Spatial" "SCT"

In SPATA2, assays are strictly sorted by molecular modality and thus, only one assay of molecular modality gene can exist. In this scenario, one has to choose the assay to use for conversion. Related matrices can always be added afterwards via addProcessedMatrix(). Once you picked the assay of interest, define the molecular modality. Optionally, you use either of ‘gene’, ‘metabolite’ or ‘protein’. Read more on how SPATA2 deals with molecular variables here.

2.4 Image name

Seurat objects contain spatial data in slot @images, which is a list that can contain multiple slots. Similarly to assays, if this list contains more than one SpatialImage objects, the one to use must be specified. Furthermore, for appropriate scaling to image resolutions the scale factor to use might need to be specified. This is usually the case for Visium experiments.

2.5 Convert

Eventually we can convert the object with asSPATA2(). Note, that we specify assay_name since the Seurat object contains two assays. However, we must not specify img_name since only one image (anterior1 is present).

# two assays 
names(brain@assays)
## [1] "Spatial" "SCT"
# one image
names(brain@images)
## [1] "anterior1"
# use assay Spatial
spata_object1 <-
 asSPATA2(
   object = brain,
   sample_name = "mouse_brain",
   platform = "VisiumSmall",
   img_scale_fct = "lowres",
   assay_name = "Spatial", 
   assay_modality = "gene"
   )

show(spata_object1)
## SPATA2 object of size: 2696 x 31053 (spots x molecules)
## Sample name: mouse_brain 
## Platform: VisiumSmall 
## Molecular assays (1):
## 1. Assay
##  Molecular modality: gene
##  Distinct molecules: 31053
##  Matrices (2):
##  -counts (active)
##  -data
## Images (1):
## - anterior1 (599x600 px, active)
## Meta variables (10): sample, orig.ident, nCount_Spatial, nFeature_Spatial, slice, region, nCount_SCT, nFeature_SCT, SCT_snn_res.0.8, seurat_clusters
# use assay SCT
spata_object2 <-
 asSPATA2(
   object = brain,
   sample_name = "mouse_brain",
   platform = "VisiumSmall",
   img_name = "anterior1",
   img_scale_fct = "lowres",
   assay_name = "SCT",
   assay_modality = "gene"
   )

show(spata_object2)
## SPATA2 object of size: 2696 x 17668 (spots x molecules)
## Sample name: mouse_brain 
## Platform: VisiumSmall 
## Molecular assays (1):
## 1. Assay
##  Molecular modality: gene
##  Distinct molecules: 17668
##  Matrices (3):
##  -counts (active)
##  -data
##  -scale.data
## Images (1):
## - anterior1 (599x600 px, active)
## Meta variables (10): sample, orig.ident, nCount_Spatial, nFeature_Spatial, slice, region, nCount_SCT, nFeature_SCT, SCT_snn_res.0.8, seurat_clusters
# left plot
SpatialPlot(brain, features = "nCount_Spatial")

# right plot
plotSurface(spata_object1, color_by = "nCount_Spatial", pt_size = 1)

3. AnnData

For demonstrating Scanpy/Squidpy compatibility, we provide a preprocessed 10X Visium dataset from the mouse brain. It was exported as h5ad file from Python using AnnData.write_h5ad().

library(anndata)
library(SPATA2)
library(tidyverse)

# required for anndata R package; path to python (with anndata installed)
reticulate::use_python("/my/path/to/python3")
# adjust to your directry
dir = "/Users/simonf/PhD/TEMP/"

curl::curl_download(
  url = "https://www.dropbox.com/s/pfpqyg1ds1d52f3/stab_wound_injury_anndata.h5ad?raw=1",
  destfile = paste0(dir,"mouse_brain_adata.h5ad")
)

h5ad files can be loaded into R by using the read_h5ad() function from the anndata R package.

adata_object <- anndata::read_h5ad(filename = paste0(dir,"mouse_brain_adata.h5ad"))

The object is converted to SPATA2 using asSPATA2().

spata_object <- asSPATA2(adata_object, sample_name = "mousebrain", platform = "VisiumSmall", modality = "gene")

If no AnnData layers are named, adata.X will be imported as normalized matrix. Custom AnnData layers can be added as either count matrix (count_mtr_name), normalized matrix (normalized_mtr_name), or scaled matrix (normalized_mtr_name) in the SPATA2 object, for instance as asSPATA2(adata, count_mtr_name = "count_matrix"). That’s it - the SPATA2 object can be used for any downstream analyses, such as spatial annotation or spatial trajectory screening.