Issues and Suggestions for Malignant Cell Fraction Analysis

[bhyu0217]

Hello,

Thank you so much for developing the fascinating tool, which has been incredibly useful for my research!

I am inferring malignant cell fractions through CNV and then performing deconvolution on the non-malignant fraction using matched sc/snRNA-seq data.
However, I seem to encounter errors in downstream analyses due to the absence of a reference profile for malignant in the SpaCET object.

Here is the error I encountered during the analysis:

SpaCET_obj <- SpaCET.deconvolution.matched.scRNAseq( spatial.obj, sc_includeMalignant = FALSE, cancerType = "GBM", sc_counts = sc_counts, sc_annotation = sc_annotation, sc_lineageTree = sc_lineageTree, sc_nCellEachLineage = 100, coreNo = 8 )

SpaCET_obj <- SpaCET.deconvolution.malignant(SpaCET_obj, Malignant = "Malignant", malignantCutoff = 0.7, coreNo = 8)

[1] "Identify 2 malignant cell states" Error in Ref$refProfiles[, knownCellTypes] : subscript out of bounds In addition: Warning messages: 1: Zero sample variances detected, have been offset away from zero 2: Zero sample variances detected, have been offset away from zero

I noticed that applying SpaCET_obj@results$deconvolution$Ref <- NULL allows the function to run, but it results in an error where all spots (including non-malignant spots) are classified into cancer cell states.

Additionally, during colocalization analysis, the "reference_rho" and "reference_pv" values associated with "malignant" are displayed as "NA" values.

SpaCET_obj <- SpaCET.CCI.colocalization(SpaCET_obj)

SpaCET_obj@results$CCI
$colocalization
cell_type_1 cell_type_2 fraction_product fraction_rho
Glial_Malignant Glial Malignant 2.510831e-02 -0.103
Vascular_Malignant Vascular Malignant 1.556394e-01 -0.702
Neuronal_Malignant Neuronal Malignant 3.049182e-02 0.001
Immune_Malignant Immune Malignant 3.716447e-02 -0.319
Astro_Malignant Astro Malignant 1.483616e-02 -0.071
Oligo_Malignant Oligo Malignant 9.181587e-03 -0.151
OPC_Malignant OPC Malignant 9.329728e-04 0.057
Endo_Malignant Endo Malignant 4.534058e-02 -0.596

fraction_pv reference_rho reference_pv
Glial_Malignant 7.997172e-05 NA NA
Vascular_Malignant 7.822941e-218 NA NA
Neuronal_Malignant 9.616220e-01 NA NA
Immune_Malignant 4.353278e-36 NA NA
Astro_Malignant 6.225376e-03 NA NA
Oligo_Malignant 5.837621e-09 NA NA
OPC_Malignant 2.986717e-02 NA NA
Endo_Malignant 1.141059e-141 NA NA

I also wanted to ask if it might be possible to implement a function that allows the deconvolution of malignant spots using malignant cell states defined in the matched sc/snRNA-seq data after inferring malignant fractions using CNV.
If this is available, it would make it much easier to directly apply cancer cell states established in matched sc/snRNA-seq data to the spatial level.

Thank you for your time and support.

[beibeiru]

Hi, @bhyu0217 ,

We are delighted that our tool can help you with your studies!

I also wanted to ask if it might be possible to implement a function that allows the deconvolution of malignant spots using malignant cell states defined in the matched sc/snRNA-seq data after inferring malignant fractions using CNV.

Do you mean your single-cell data includes malignant cells? If yes, you can set sc_includeMalignant=TRUE to skip inferring malignant fractions using CNV, and decovolve directly with your matched reference, just like https://data2intelligence.github.io/SpaCET/articles/oldST_PDAC.html.

If I misunderstand you or the previous method does not work, please share your spatial.obj, sc_counts, sc_annotation, and sc_lineageTree with me. I will fix it soon.

Best,
Beibei

[bhyu0217]

Hi @beibeiru,

Thank you for your quick and thoughtful response.
I apologize if my question was unclear.

After inferring malignant cell fractions using CNV, for spots where the malignant fraction exceeds a certain threshold (e.g., > 0.7), I would like to perform deconvolution of malignant fractions using the matched single-cell data.

For example, if my single-cell data defines multiple malignant states such as "Malignant A", "Malignant B", and "Malignant C", I'd like to deconvolve these states within the malignant spots inferred by CNV.

CNV-based malignant fraction inference -> malignant state deconvolution using matched single-cell data
(similar to how normal cell fractions are deconvolved into major lineages and sub-lineages)

I have tried using sc_includeMalignant=TRUE for malignant cell deconvolution, as you suggested, but in my datasets, starting with CNV-based malignant fraction inference appears to yield more accurate results.

Does this clarify my question?
Thank you.

Best,
Bohyeon

[beibeiru]

Hi, Bohyeon,

It is very clear right now. Thanks.
I will let you know once I address your issue. Maybe 1 or 2 days.

Best,
Beibei

[beibeiru]

Hi, Bohyeon,

I have added a new function SpaCET.deconvolution.malignant.customized.scRNAseq which can meet your requirements. Just reinstall our package to use it. Please see the following example for more details.

library(SpaCET)

oldST_Path <- system.file("extdata", 'oldST_PDAC', package = 'SpaCET')

# load ST data to create a SpaCET object.
load(paste0(oldST_Path,"/st_PDAC.rda"))

SpaCET_obj <- create.SpaCET.object(
  counts=counts,
  spotCoordinates=spotCoordinates,
  imagePath=NA,
  platform = "oldST"
)

# load scRNA-Seq data 
load(paste0(oldST_Path,"/sc_PDAC.rda"))

# separate scRNA-Seq data into two groups, cancer and normal.
sc_lineageTree_cancer <- sc_lineageTree[1]
sc_lineageTree_normal <- sc_lineageTree[2:14]

sc_annotation_cancer <- sc_annotation[sc_annotation[,"bio_celltype"]%in%c("Cancer clone A","Cancer clone B"),]
sc_annotation_normal <- sc_annotation[!sc_annotation[,"bio_celltype"]%in%c("Cancer clone A","Cancer clone B"),]

sc_counts_cancer <- sc_counts[,sc_annotation_cancer[,"cellID"]]
sc_counts_normal <- sc_counts[,sc_annotation_normal[,"cellID"]]

# infer malignant fraction with CNV signature and deconvolve normal fraction with matched normal scRNAseq.
SpaCET_obj <- SpaCET.deconvolution.matched.scRNAseq(
  SpaCET_obj, 
  sc_includeMalignant=FALSE,
  cancerType="PAAD",
  sc_counts=sc_counts_normal, 
  sc_annotation=sc_annotation_normal, 
  sc_lineageTree=sc_lineageTree_normal, 
  coreNo=8
)

# deconvolve malignant fraction with customized tumor scRNAseq.
SpaCET_obj <- SpaCET.deconvolution.malignant.customized.scRNAseq(
  SpaCET_obj, 
  Malignant="Malignant", 
  sc_counts=sc_counts_cancer, 
  sc_annotation=sc_annotation_cancer, 
  sc_lineageTree=sc_lineageTree_cancer, 
  sc_nCellEachLineage=100, 
  coreNo=8
)

Please let me know whether it works for you!

Best,
Beibei