Integration for Vizgen data

[yous0614]

Hi team,

I am currently using GiottoSuite and I found the package smooth and amazing, I really like it.
However, I encounter a problem for integration of a couple of Vizgen datasets.
My goal is to integrate 10 human brain slides and at least to identify neurons, microglia, astrocytes cell types etc.

I was able to run the whole [vizgen brain] tutorial, however, the integration for the vizgen datasets isn't provided.
I then try to combine them using joinGiottoObjects() as in [Visium Prostate Integration] tag.
I was able to create a great align panel like this.

However, when I try to move on to filter or normalizeGiotto(), I found the error message like:

Error in matrixStats::colSums2(temp_matrix) :
  Argument 'x' must be of type logical, integer or numeric, not 'list'.

I wonder is there any solution for combine and analyse these dataset?

Shih-Feng
PhD student @ washU

[RubD]

Hi @yous0614 , can you provide a bit more detail? What's the exact script that you ran? It seems like the joint object doesn't have a raw count matrix. Did you use calculateOverlapRaster to identify which transcripts are within which cells (segmentations) and overlapToMatrix to aggregate the overlap information into a matrix?

[yous0614]

Thanks for the speedy replies.
Yes, so I followed mostly the tutorial on this page to prepare my giotto object. (https://giottosuite.readthedocs.io/en/latest/subsections/datasets/vizgen_mouse_brain.html),
I created the objects by createGiottoObject() and addCellMetadata(), then attached and update figure info on all the staining images by createGiottoLargeImageList(), addGiottoImage() and updateGiottoLargeImage(). Filtered, normalized, PCA, UMAP, NearestNetwork, LeidenCluster to generate the figure for each object.

The tutorial doesn't mentioned how to integrate, I then run a for loop to make 10 objects in a list and join them by another tutorial on (https://giottosuite.readthedocs.io/en/latest/subsections/datasets/visium_prostate_integration.html) using:

vizgen_integrated = joinGiottoObjects(gobject_list = vizgen_objects,
                                      gobject_names = gobject_names, x_shift = c(seq(0, 15000*9, 15000)),
                                      join_method = 'shift', x_padding = 1000)

That's how I made the graph above, however, as you mentioned calculateOverlapRaster and overlapToMatrix, I went back and check vizgen@spatial_info, I realized I have to add the polygon and gpoint onto the file as the 'subcellular' section in the first tutorial.

createGiottoPoints() and addGiottoPoints() was used to make point data and attach.
readPolygonFilesVizgen() can't be used since H5Fopen_flags = "H5F_ACC_RDONLY" isn't an option, I used readPolygonFilesVizgenHDF5() then addGiottoPolygons() and addSpatialCentroidLocations() to attach the polygons.

By then, I can pass through calculateOverlapRaster() and overlapToMatrix(), however, I realized the spatial info in Vizgen object is stored in $z0, $z1, $z2, $z3, ...$z6, and though Slot "name", "spatVector", "spatVectorCentroids" are updated in every layer, the "overlaps" only updated in $z0, so I have to do poly_info= 'z0' to use it.

However, I still got

> vizgen_integrated = normalizeGiotto(vizgen_integrated)
Error in matrixStats::colSums2(temp_matrix) :
  Argument 'x' must be of type logical, integer or numeric, not 'list'.

10 samples is too big so I tried to merge the first sample itself as test drive.
Here I attached the showGiottoExpression(), showGiottoSpatialInfo() and showGiottoCellMetadata() for before (a giotto object name vizgen) and after (merged object named vizgen_integrated) joinning object.
Hope that can help. Thank you.

vizgen and vizgen_integrated.txt

[yous0614]

Update:
I figured perhaps the z0-z6 could be the problem, I tried to attach only one z-stack by readPolygonFilesVizgenHDF5(), along with gpoint data, I can perform calculateOverlapRaster() and overlapToMatrix(..., poly_info = 'z3'). By specify that, I can perform normalization on combined join Giotto object. But it warns:

> normalizeGiotto(combo1, spat_unit= 'z3', feat_type= 'rna')

 first scale feats and then cells
An object of class giotto
Steps and parameters used:

$`0_normalize`
          gobject         spat_unit         feat_type expression_values
         "combo1"              "z3"             "rna"             "raw"
     norm_methods library_size_norm       scalefactor          log_norm
       "standard"            "TRUE"            "6000"            "TRUE"
       log_offset           logbase       scale_feats       scale_cells
              "1"               "2"            "TRUE"            "TRUE"
      scale_order             theta       update_slot           verbose
    "first_feats"             "100"          "scaled"            "TRUE"

Warning message:
In libNorm_giotto(mymatrix = raw_expr[], scalefactor = scalefactor) :
  Total library size or counts for individual spat units are 0.
 This will likely result in normalization problems.
 filter (filterGiotto) or impute (imputeGiotto) spatial units.

I guess there should be a way to make a bypass by attaching HDF5 polygons as default in Slot "cell" but not "z3"?
Best,
Shih-Feng

[RubD]

Hi Shih-Feng, we're currently behind on our tutorials and examples, but we already have a working solution for dealing with multiple stacks, see the aggregateStacks function in suite or suite_dev (dev branch). We hope to update some of our tutorials to showcase it's usage.

[yous0614]

Hi Dr. Dries,
thanks to point that out. I am not very familiar with the giotto object structure, so I tried to implement it by following one of the tutorial I found. (https://github.com/drieslab/GiottoData/blob/b22ff768159b4173f29dadfecd86cb176593ca95/inst/Mini_datasets/Vizgen/mini_vizgen_script.R)

Here I tried to combine the vizgen object twice by inserting the joinGiottoObjects() after step 2 in different ways, I hope you can correct me that which way is more logical, or is there any parameter I set it wrong.

1. joinGiottoObjects() -> calculateOverlapRaster() -> overlapToMatrix() -> aggregateStacks() -> filterGiotto() -> normalizeGiotto()

testcombo1 = joinGiottoObjects(c(vizgen, vizgen), c('v1', 'v2'), join_method= 'shift', x_shift = c(0, 15000), x_padding= 1000)
testcombo1 = calculateOverlapRaster(testcombo1, spatial_info= 'z0', feat_info= 'rna', feat_subset_column= 'global_z', feat_subset_ids= 0)
testcombo1 = overlapToMatrix(testcombo1, poly_info = 'z0', feat_info = 'rna', name = 'raw')
testcombo1 = calculateOverlapRaster(testcombo1, spatial_info= 'z1', feat_info= 'rna', feat_subset_column= 'global_z', feat_subset_ids= 1)
testcombo1 = overlapToMatrix(testcombo1, poly_info = 'z1', feat_info = 'rna', name = 'raw')
testcombo1 = aggregateStacks(testcombo1, spat_units= c('z0', 'z1'), feat_type= 'rna', values= 'raw', summarize_expression= 'sum', summarize_locations= 'mean', new_spat_unit= 'aggregate')
testcombo1 = filterGiotto(testcombo1, spat_unit= 'aggregate', expression_threshold = 1, feat_det_in_min_cells = 3, min_det_feats_per_cell = 5, expression_values= 'raw', poly_info= 'z0')

completed 1: preparation
Error in S4_expr@exprMat[filter_bool_feats, ] :
incorrect number of dimensions

2.calculateOverlapRaster() -> overlapToMatrix() -> aggregateStacks()->joinGiottoObjects() -> filterGiotto() -> normalizeGiotto()

vizgen = calculateOverlapRaster(vizgen, spatial_info= 'z0', feat_info= 'rna', feat_subset_column= 'global_z', feat_subset_ids= 0)
vizgen = overlapToMatrix(vizgen, poly_info = 'z0', feat_info = 'rna', name = 'raw')
vizgen = calculateOverlapRaster(vizgen, spatial_info= 'z1', feat_info= 'rna', feat_subset_column= 'global_z', feat_subset_ids= 1)
vizgen = overlapToMatrix(vizgen, poly_info = 'z1', feat_info = 'rna', name = 'raw')
vizgen = aggregateStacks(vizgen, spat_units= c('z0', 'z1'), feat_type= 'rna', values= 'raw', summarize_expression= 'sum', summarize_locations= 'mean', new_spat_unit= 'aggregate')
testcombo2 = joinGiottoObjects(c(vizgen, vizgen), c('v1', 'v2'), join_method= 'shift', x_shift = c(0, 15000), x_padding= 1000)

Join method:shift
start updating objects
A) Update giotto Objects
0. Update cell and feature IDs
1. Update expression IDs
Error in dimnamesGets(x, value) :
length of Dimnames[[2]] (32271) is not equal to Dim[2] (32688)

3.filterGiotto() -> joinGiottoObjects() -> calculateOverlapRaster() -> overlapToMatrix() -> aggregateStacks() -> normalizeGiotto()

vizgen = filterGiotto(gobject = vizgen, expression_threshold = 1, feat_det_in_min_cells = 100, min_det_feats_per_cell = 20)
testcombo3 = joinGiottoObjects(c(vizgen, vizgen), c('v1', 'v2'), join_method= 'shift', x_shift = c(0, 15000), x_padding= 1000)
testcombo3 = calculateOverlapRaster(testcombo3, spatial_info= 'z0', feat_info= 'rna', feat_subset_column= 'global_z', feat_subset_ids= 0)
testcombo3 = overlapToMatrix(testcombo3, poly_info = 'z0', feat_info = 'rna', name = 'raw')
testcombo3 = calculateOverlapRaster(testcombo3, spatial_info= 'z1', feat_info= 'rna', feat_subset_column= 'global_z', feat_subset_ids= 1)
testcombo3 = overlapToMatrix(testcombo3, poly_info = 'z1', feat_info = 'rna', name = 'raw')
testcombo3 = aggregateStacks(testcombo3, spat_units= c('z0', 'z1'), feat_type= 'rna', values= 'raw', summarize_expression= 'sum', summarize_locations= 'mean', new_spat_unit= 'aggregate')
testcombo3 = normalizeGiotto(testcombo3, spat_unit= 'aggregate', scalefactor= 5000, verbose=T)

Error in x$.self$finalize() : attempt to apply non-function
In addition: Warning message:
In libNorm_giotto(mymatrix = raw_expr[], scalefactor = scalefactor) :
Total library size or counts for individual spat units are 0.
This will likely result in normalization problems.
filter (filterGiotto) or impute (imputeGiotto) spatial units.
Error in x$.self$finalize() : attempt to apply non-function
  
first scale feats and then cells
Error in x$.self$finalize() : attempt to apply non-function
Error in x$.self$finalize() : attempt to apply non-function

[yous0614]

Update:

Thanks for the reply.
I found my problem, when use joinGiottoObjects(), in object@expression it generates not only a cell-by-gene $matrix, but also a list of genes called $sort_all_feats. The $Matrix and $sort_all_feats in object@expression$cell$rna$raw are considered a list, therefore it causes a problem when get_expression_values() trying to get a matrix from vizgen object and put forward in filterGiotto(), which causes:

Error in rowSums_flex(expr_values >= expression_threshold) :
  'list' object cannot be coerced to type 'double'

I bypassed the problem by

vizgen_integrated@expression$cell$rna$raw = 
get_expression_values(vizgen_integrated, spat_unit= 'cell', feat_type= 'rna', output = 'matrix', values= 'raw')[[1]]

Thanks