How to Test and Store Multiple Parameters or Analyses?

The default Giotto workflow is similar to other scRNA-seq workflows and does not require you to provide a custom name for each analysis (e.g. PCA, UMAP, …), but running an analysis twice will overwrite the previous results with a warning.

However, there are situations where being able to run and store multiple analyses can be advantageous:

  • Test multiple parameters for a single analysis

  • Test multiple combinations across functions (See Example: hvg->pca->umap)

  • Use different output results as input for downstream analyses (See Example: spatial genes)

Multiple Analysis

We will use the seqFish+ somatosensory cortex as an example dataset after creating and processing a Giotto object.

1. Calculate Highly Variable Genes (2 Methods)

# using the loess method
VC_test <- calculateHVG(gobject = VC_test,
            method = 'cov_loess', difference_in_cov = 0.1,
            HVGname = 'loess_hvg')
Loess 
# using the expression groups method
VC_test <- calculateHVG(gobject = VC_test
            , method = 'cov_group', zscore_threshold = 1,
            HVGname = 'group_hvg')
Group 
# compare the highly variable genes between two methods
gene_metadata = fDataDT(VC_test)
mytable = table(loess = gene_metadata$loess_hvg, group = gene_metadata$group_hvg)
Group

2. Perform Multiple PCAs

  • Using the 2 different HVG sets (loess_genes and group_genes)

  • Store PCA results using custom names (‘pca_loess’ and ‘pca_group’)

  • Plot PCA results

## 2. PCA ##
# pca with genes from loess
loess_genes = gene_metadata[loess_hvg == 'yes']$gene_ID
VC_test <- runPCA(gobject = VC_test, genes_to_use = loess_genes, name = 'pca_loess', scale_unit = F)
plotPCA(gobject = VC_test, dim_reduction_name = 'pca_loess')
Group 
# pca with genes from group
group_genes = gene_metadata[group_hvg == 'yes']$gene_ID
VC_test <- runPCA(gobject = VC_test, genes_to_use = group_genes, name = 'pca_group', scale_unit = F)
plotPCA(gobject = VC_test, dim_reduction_name = 'pca_group')
Group

3. Create Multiple UMAPs

  • Using the 2 different PCA results (‘pca_loess’ and ‘pca_group’)

  • Store UMAP results using custom names (‘umap_loess’ and ‘umap_group’)

  • Plot UMAP results

## 3. UMAP ##
  VC_test <- runUMAP(VC_test, dim_reduction_to_use = 'pca', dim_reduction_name = 'pca_loess',
         name = 'umap_loess', dimensions_to_use = 1:30)
  plotUMAP(gobject = VC_test, dim_reduction_name = 'umap_loess')
Group 
VC_test <- runUMAP(VC_test, dim_reduction_to_use = 'pca', dim_reduction_name = 'pca_group',
           name = 'umap_group', dimensions_to_use = 1:30)
plotUMAP(gobject = VC_test, dim_reduction_name = 'umap_group')
Group

4. Create Multiple Spatial Networks

  • Create spatial with multiple k’s and other parameters (k=5, k=10, k=100 & maximum_distance=200)

  • Subset field 1

  • Visualize network on field 1 (‘spatial_network’, ‘large_network’, ‘distance_work’)

Spatial Network

## 4. spatial network
VC_test <- createSpatialNetwork(gobject = VC_test, method = 'kNN', k = 5) # standard name: 'spatial_network'
VC_test <- createSpatialNetwork(gobject = VC_test, method = 'kNN', k = 10, name = 'large_network')  VC_test <- createSpatialNetwork(gobject = VC_test, method = 'kNN', k = 100, maximum_distance_knn = 200, minimum_k = 2, name = 'distance_network')

## visualize different spatial networks on first field (~ layer 1)
cell_metadata = pDataDT(VC_test)
field1_ids = cell_metadata[Field_of_View == 0]$cell_ID
subVC_test = subsetGiotto(VC_test, cell_ids = field1_ids)

spatPlot(gobject = subVC_test, show_network = T,
     network_color = 'blue', spatial_network_name = 'spatial_network')
Group

Large Network

spatPlot(gobject = subVC_test, show_network = T,
     network_color = 'blue', spatial_network_name = 'large_network')
Group

Distance Network

spatPlot(gobject = subVC_test, show_network = T,
     network_color = 'blue', spatial_network_name = 'distance_network')
Group

5. Find Spatial Genes (Multiple Methods)

  • Use the different spatial networks as input to identify spatial genes with the rank method

  • Visualize top spatial genes for 2 methods

Large Network Spatial Genes

## 5. spatial genes
 # the provided spatial_network_name can be given to downstream analyses

 # spatial genes based on large network
 ranktest_large = binSpect(VC_test,
               subset_genes = loess_genes,
               bin_method = 'rank',
               spatial_network_name = 'large_network')


 spatGenePlot(VC_test,
          expression_values = 'scaled',
          genes = ranktest_large$genes[1:6], cow_n_col = 2, point_size = 1,
          genes_high_color = 'red', genes_mid_color = 'white', genes_low_color = 'darkblue', midpoint = 0)
Group

Distance Network Spatial Genes

# spatial genes based on distance network
ranktest_dist = binSpect(VC_test,
             subset_genes = loess_genes,
             bin_method = 'rank',
             spatial_network_name = 'distance_network')
spatGenePlot(VC_test,
         expression_values = 'scaled',
         genes = ranktest_dist$genes[1:6], cow_n_col = 2, point_size = 1,
         genes_high_color = 'red', genes_mid_color = 'white', genes_low_color = 'darkblue', midpoint = 0)
Group