predictTilState ERROR

[france-hub]

Hi!

Thanks for your package.

I am trying to follow the following case study https://carmonalab.github.io/ProjecTILs_CaseStudies/SadeFeldman_ortho.html using my scRNAseq data (CD3+ T cells).

However, when I run make.projection on my Seurat object I get the following error:

Error in predictTilState(sce, human = human) : Too many genes not found In addition: Warning message: In predictTilState(sce, human = human) : The following genes were not found in the dataset provided CD2,CD3D,CD3E,CD247,LCK,CD8B,CD8A,CD4,...

When I run: c("CD3D", "CD3E", "CD247", "LCK") %in% row.names(sobj) TRUE TRUE TRUE TRUE

sobj here is my seurat object.

I noticed that in the query_example_seurat the gene name format is with lower cases (eg "Cd3d"). However even if I change the format of my gene names I get the same error.

Could you please help me?

Thanks a lot

Francesco

[mass-a]

Hello Francesco,

by the name of your genes I guess you have human data. Can you confirm that you are running make.projection with the human.ortho=TRUE option?

If that is the case, in which assay of the seurat object are the data stored? and do you have a "data" or "counts" slot for that assay?

[france-hub]

Thank you for your reply.

Yes, I am running query.projected <- make.projection(sobj, ref = ref, human.ortho = T)

The assay is RNA and I have "data" and "counts" slots in that assay.

Thank you again Francesco

[mass-a]

It seems that TILPRED doesn't get the data in the right format, for some reason.

Are the data correctly converted to the sce format if your run the following commands?

dim(sobj@assays$RNA@counts)
dim(sobj@assays$RNA@data)
sce <- as.SingleCellExperiment(sobj)
dim(sce)
c("CD3D", "CD3E", "CD247", "LCK") %in% row.names(sce)

and then

sce.pred <- predictTilState(sce, human=TRUE)

What version of TILPRED do you have?

packageVersion("TILPRED")

[france-hub]

This is the output:

`> dim(sobj@assays$RNA@counts) [1] 8781 26592

dim(sobj@assays$RNA@data) [1] 8781 26592 sce <- as.SingleCellExperiment(sobj) dim(sce) [1] 8781 26592 c("CD3D", "CD3E", "CD247", "LCK") %in% row.names(sce) [1] TRUE TRUE TRUE TRUE sce.pred <- predictTilState(sce, human=TRUE) Error in predictTilState(sce, human = TRUE) : Too many genes not found In addition: Warning message: In predictTilState(sce, human = TRUE) : The following genes were not found in the dataset provided CD19,NAPSB,CD22,NCF1C,IGLL5,VPREB3,ADAM28,CIITA,FCRL1,HLA-DOB,NCF1B,LY86,ELK2AP,CR2,PKIG,CYBASC3,IGJ,MFAP5,FBLN1,COL1A2,BGN,DCN,COL1A1,COL3A1,SPARC,C1S,LUM,MMP2,THY1,PCOLCE,PMP22,SFRP4,MGP,FSTL1,C3,C1R,SERPING1,EFEMP1,FN1,NNMT,FBLN2,CTSK,DPT,SFRP2,CXCL12,PPIC,CXCL14,RARRES2,MFAP4,WISP2,COMP,CTHRC1,CTGF,PLA2G2A,COL8A1,TAGLN,POSTN,PTRF,GNG11,CALD1,S100A16,CNN3,TM4SF1,A2M,EGFL7,HYAL2,RAMP2,TFPI,CYR61,RAMP3,AQP1,SPARCL1,C10orf10,RNASE1,FABP4,CCL14,DARC,CCL21,CD14,FAM26F,TMEM176B,CPVL,GPX1,CTSL1,C1QA,C1QB,C1QC,VSIG4,APOC1,SPP1,APOE,IL8,SEPP1,CXCL10,S100A13,PLP1,MLANA,PRAME,TYR,APOD,PMEL,MIA,QPCT,GDF15,DCT,APOC2,SERPINA3,TYRP1 . Doesn't look too bad but prediction performance might be affected. packageVersion("TILPRED") [1] ‘1.0.1’`

[mass-a]

Ok I see now. Too many genes are missing and TILPRED fails to score its signatures.

I think you have two options: 1) either you include more genes from the expression matrix (currently you have 8781 genes, I guess some were filtered out?)

2) or, if you are sure that you have only T cells (you mentioned your data are CD3+), you could simply run ProjecTILs without the TILPRED signature filter:

query.projected <- make.projection(sobj, ref = ref, human.ortho = T, filter.cells=FALSE)

I hope this helps -massimo

[france-hub]

Thank you! I have tried your second option, but it seems that still too many genes are missing:

query.projected` <- make.projection(sobj, ref = ref, human.ortho = T, filter.cells = FALSE, skip.normalize = TRUE)
[1] "Using assay integrated for query"
[1] "Transforming expression matrix into space of mouse orthologs"
Error in make.projection(sobj, ref = ref, human.ortho = T, filter.cells = FALSE,  : 
  Too many genes missing. Check input object format

Thank you for your help and patience Francesco

[mass-a]

From your output, I see your are using the "integrated" assay for the query.

Is that what you want, and in that case how many genes do you have in that assay? dim(sobj@assays$integrated@data)

Otherwise you should switch to the "RNA" assay, as seemed to be your intention from a previous comment.

[france-hub]

Thank you very much for your suggestions. Apologize for the delayed reply

Francesco