PRECAST: simulation

Prepare the PRECASTObject with preprocessing step.

Next, we use CreatePRECASTObject() to create a PRECASTObject based on the Seurat list object seuList. This function will do three things:

• 1. Filter low-quality spots and genes, controlled by the arguments premin.features and premin.spots, respectively; the spots are retained in raw data (seuList) with at least premin.features number of nonzero-count features (genes), and the genes are retained in raw data (seuList) with at least premin.spots number of spots. To ease presentation, we denote the filtered Seurat list object as data_filter1.
• 2. Select the top 2,000 variable genes (by setting gene.number=2000) for each data batch using FindSVGs() function in DR.SC package for spatially variable genes or FindVariableFeatures() function in Seurat package for highly variable genes. Next, we prioritized genes based on the number of times they were selected as variable genes in all samples and chose the top 2,000 genes. Then denote the Seurat list object as data_filter2, where only 2,000 genes are retained.
• 3. Conduct strict quality control for data_filter2 by filtering spots and genes, controlled by the arguments postmin.features and postmin.spots, respectively; the spots are retained with at least post.features nonzero counts across genes; the features (genes) are retained with at least postmin.spots number of nonzero-count spots. Usually, no genes are filltered because these genes are variable genes.

If the argument customGenelist is not NULL, then this function only does (3) based on customGenelist gene list.

In this simulated dataset, we don’t require to select genes, thus, we set customGenelist=row.names(seuList[[1]]), representing the user-defined gene list. User can retain the raw seurat list object by setting rawData.preserve = TRUE.

## Create PRECASTObject
set.seed(2022)
PRECASTObj <-  CreatePRECASTObject(seuList, customGenelist=row.names(seuList[[1]]))

## User can retain the raw seuList by the following commond.
##  PRECASTObj <-  CreatePRECASTObject(seuList, customGenelist=row.names(seuList[[1]]), rawData.preserve = TRUE)

Add the model setting

Add adjacency matrix list and parameter setting of PRECAST. More model setting parameters can be found in model_set().

## check the number of genes/features after filtering step
PRECASTObj@seulist

## seuList is null since the default value `rawData.preserve` is FALSE.
PRECASTObj@seuList

## Add adjacency matrix list for a PRECASTObj object to prepare for PRECAST model fitting.
PRECASTObj <-  AddAdjList(PRECASTObj, platform = "Visium")

## Add a model setting in advance for a PRECASTObj object: verbose =TRUE helps outputing the information in the algorithm; coreNum set the how many cores are used in PRECAST. If you run PRECAST for multiple number of clusters, you can set multiple cores; otherwise, set it to 1. 
PRECASTObj <- AddParSetting(PRECASTObj, Sigma_equal=FALSE, maxIter=30, verbose=TRUE,
                             coreNum =1)

Fit PRECAST

For function PRECAST, users can specify the number of clusters \(K\) or set K to be an integer vector by using modified BIC(MBIC) to determine \(K\). For convenience, we give a single K here.

### Given K
set.seed(2022)
PRECASTObj <- PRECAST(PRECASTObj, K=7)

## Reset  parameters by increasing cores.
PRECASTObj2 <- AddParSetting(PRECASTObj, Sigma_equal=FALSE, maxIter=30, verbose=TRUE,
                             coreNum =2)
set.seed(2023)
PRECASTObj2 <- PRECAST(PRECASTObj2, K=6:7)

resList2 <- PRECASTObj2@resList
PRECASTObj2 <- SelectModel(PRECASTObj2)

Select a best model and re-organize the results by useing SelectModel(). Even though K is not a vector, it is also necessary to run SelectModel() to re-organize the results in PRECASTObj. The selected best K is 7 by using command str(PRECASTObj@resList).

## check the fitted results: there are four list for the fitted results of each K (6:9).
str(PRECASTObj@resList)
## backup the fitted results in resList
resList <- PRECASTObj@resList
# PRECASTObj@resList <- resList
PRECASTObj <- SelectModel(PRECASTObj)
## check the best and re-organized results
str(PRECASTObj@resList) ## The selected best K is 7

Use ARI to check the performance of clustering:

true_cluster <- lapply(PRECASTObj@seulist, function(x) x$true_cluster)
str(true_cluster)
mclust::adjustedRandIndex(unlist(PRECASTObj@resList$cluster), unlist(true_cluster))

We provide two methods to correct the batch effects in gene expression level. Method (1) is using only PRECAST results to obtain the batch corrected gene expressions if the species of data is unknown or the number of overlapped housekeeping genes between the variable genes in PRECASTObj@seulist and the genes in database is less than five. Method (2) is using bouth housekeeping gene and PRECAST results to obtain the batch corrected gene expressions.

• Note: to obtain batch corrected gene expressions based on housekeeping genes as the negative control, users must specify the species of data source and use gene symbol names in PRECASTObj@seulist.

Integrate the two samples by the function IntegrateSpaData. Because this is a simulated data, we use Method (1) by setting species='unknown'.

seuInt <- IntegrateSpaData(PRECASTObj, species='unknown')
seuInt 
## The low-dimensional embeddings obtained by PRECAST are saved in PRECAST reduction slot.