deconvolute doesn't work with custom_1 markers

[alievakrash]

Hello,

I am trying deconvolution of bulk RNA-seq data based on brain structure expression patterns on my organoids. It works great when I use structures from custom_2, but when I run it on custom_1 or _3 I get the following message with the deconvolute function:

Error in intI(j, n = x@Dim[2], dn[[2]], give.dn = FALSE) : invalid character indexing

Do you know how to solve this issue?

Cheers, Maria

[joschif]

Hi Maria,

for me the function works as expected. Can you maybe provide me with the full function call and the version of VoxHunt you are running?

Best, Jonas

[alievakrash]

Hi Jonas,

I am using R version 4.0.5 and voxhunt_0.9.2. Indeed I tried it with the example_pseudobulk dataset and it works, but not with mine... I tried to convert it to a sparsematrix but doesn;t work. Here is the code that I use and attached en example of my dataset. The counts are random, just the genes are the ones that I have in my dataset (https://we.tl/t-Jhdza3SN67).

library(voxhunt) library(tidyverse) library(RColorBrewer) library(DESeq2) library(reshape2)

load the test dataset

test<-readRDS("test_dataset.rds")

load_aba_data('

voxhunt_data')

markers <- structure_markers('E11', annotation_level = 'custom_1') %>% filter(gene%in%rownames(test))

involve_regions <- unique(markers$group)

top15 <- markers %>% filter(group%in%involve_regions) %>% group_by(group) %>% top_n(15, auc) %>% {unique(.$gene)}

top50 <- markers %>% filter(group%in%involve_regions) %>% group_by(group) %>% top_n(50, auc) %>% {unique(.$gene)}

prop_df_Age7 <- deconvolute( test[top50, ], top15, involve_regions = involve_regions, pseudo_tpm = T )

I get this error with custom_1 or custom_3

Error in intI(j, n = x@Dim[2], dn[[2]], give.dn = FALSE) : invalid character indexing

Cheers,

Maria

[joschif]

Hi Maria, thanks for the very detailed description and sorry for the late reply. It seems to me like in the above code, involve_region contains the custom_1 annotation, but the default in deconvolute() is custom_2. If you tell deconvolute() to use custom_1 by setting the argument annotation_level it should work. E.g for me the following works:

prop_df_Age7 <- deconvolute(
    test_case[top50, ], top15,
    annotation_level = 'custom_1',
    involve_regions = involve_regions,
    pseudo_tpm = T
)

Alternatively you can also omit the involve_regions argument and all regions will be used per default:

prop_df_Age7 <- deconvolute(
    test_case[top50, ], top15,
    pseudo_tpm = T
)

Hope this works for you, let me know :)

Cheers, Jonas

[alievakrash]

Hey Jonas,

Thanks for checking it. Your solution worked, but only with custom_1, not custom_3... I wasalso wondering: I want to implement this on another dataset. I was wondering how feasable it is. What format should I provide this sRNA seq dataset?

Cheers,

Maria

El mié., 12 may. 2021 17:48, jonas @.***> escribió:

Hi Maria, thanks for the very detailed description and sorry for the late reply. It seems to me like in the above code, involve_region contains the custom_1 annotation, but the default in deconvolute() is custom_2. If you tell deconvolute() to use custom_1 by setting the argument annotation_level it should work. E.g for me the following works:

prop_df_Age7 <- deconvolute( test_case[top50, ], top15, annotation_level = 'custom_1', involve_regions = involve_regions, pseudo_tpm = T )

Alternatively you can also omit the involve_regions argument and all regions will be used per default:

prop_df_Age7 <- deconvolute( test_case[top50, ], top15, pseudo_tpm = T )

Hope this works for you, let me know :)

Cheers, Jonas

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/quadbiolab/VoxHunt/issues/9#issuecomment-839880978, or unsubscribe https://github.com/notifications/unsubscribe-auth/AQN3NZPRGBX3N3GYUF27UF3TNKPLPANCNFSM44CLN52A .

[joschif]

Hi Maria,

for me any annotation level works. Maybe check if you are using the correct regions in involve_regions or just omit the argument entirely so that all regions are considered.

Cheers, Jonas

[joschif]

I wasalso wondering: I want to implement this on another dataset. I was wondering how feasable it is. What format should I provide this sRNA seq dataset?

If you want to use another scRNAseq dataset as the reference, it might be better to directly use a dedicated deconvolution tool like CIBERSORT or EPIC, which the deconvolution in VoxHunt is based on.

[alievakrash]

Thanks!

El dom, 16 may 2021 a las 20:25, jonas @.***>) escribió:

I wasalso wondering: I want to implement this on another dataset. I was wondering how feasable it is. What format should I provide this sRNA seq dataset?

If you want to use another scRNAseq dataset as the reference, it might be better to directly use a dedicated deconvolution tool like CIBERSORT https://cibersortx.stanford.edu/ or EPIC https://github.com/GfellerLab/EPIC, which the deconvolution in VoxHunt is based on.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/quadbiolab/VoxHunt/issues/9#issuecomment-841856439, or unsubscribe https://github.com/notifications/unsubscribe-auth/AQN3NZJQBI4T4DKDBCOIHHDTOAE3FANCNFSM44CLN52A .

--

Maria Alieva, PhD

Senior Postdoctoral researcher, Rios group

Princess Máxima Center for Pediatric Oncology|