.github/preparing_data.md

In shooshtarilab/TMExplorer: A Collection of Tumour Microenvironment Single-cell RNA Sequencing Datasets and Corresponding Metadata

Preparing Data

Erik Christensen 24/03/2021

Preparing Datasets For Upload

This is document is a brief example on how data can be prepared for upload to TMExplorer. I’ll use GSE75688 since it has additional columns in the matrix and a separate cell-type info file.

First let’s look at the dataset in TMExplorer.

gse75688 <- queryTME(geo_accession = 'GSE75688')[[1]]
dim(counts(gse75688))

## [1] 57915   563

counts(gse75688)[1:5,1:4]

##                             BC01_Pooled BC01_Tumor BC02_Pooled BC03_Pooled
## TSPAN6_ENSG00000000003.10          2.33       1.25       43.96        7.64
## TNMD_ENSG00000000005.5             0.00       0.00        0.00        0.00
## DPM1_ENSG00000000419.8            60.70      28.44       74.73       41.41
## SCYL3_ENSG00000000457.9           47.93       4.43        9.89        7.61
## C1orf112_ENSG00000000460.12        4.79       1.67       10.87        0.92

The counts file is a 57915x563 matrix with a single column of rownames. This is the target format. Let’s compare that to the dataset downloaded from GEO.

geo <- read.csv('GSE75688_GEO_processed_Breast_Cancer_raw_TPM_matrix.txt.gz',
                sep='\t')
dim(geo)

## [1] 57915   566

geo[1:5,1:4]

##              gene_id gene_name      gene_type BC01_Pooled
## 1 ENSG00000000003.10    TSPAN6 protein_coding        2.33
## 2  ENSG00000000005.5      TNMD protein_coding        0.00
## 3  ENSG00000000419.8      DPM1 protein_coding       60.70
## 4  ENSG00000000457.9     SCYL3 protein_coding       47.93
## 5 ENSG00000000460.12  C1orf112 protein_coding        4.79

The GEO matrix has three additional columns. In order to match the target format, we’ll drop the gene_type column, merge the gene_id and gene_name columns, then set the merged result as the rownames. NOTE: The rownames here can be gene names, gene IDs, some combination of the two like we have here, but they must be unique (R cannot set non-unique rownames) and identify the gene (instead of just being numbers).

geo$gene_type <- NULL
geo$gene_id <- paste(geo$gene_name, geo$gene_id,sep='_')
rownames(geo) <- geo$gene_id
geo$gene_id <- NULL
geo$gene_name <- NULL

dim(geo)

## [1] 57915   563

geo[1:5,1:4]

##                             BC01_Pooled BC01_Tumor BC02_Pooled BC03_Pooled
## TSPAN6_ENSG00000000003.10          2.33       1.25       43.96        7.64
## TNMD_ENSG00000000005.5             0.00       0.00        0.00        0.00
## DPM1_ENSG00000000419.8            60.70      28.44       74.73       41.41
## SCYL3_ENSG00000000457.9           47.93       4.43        9.89        7.61
## C1orf112_ENSG00000000460.12        4.79       1.67       10.87        0.92

all(geo == counts(gse75688))

## [1] TRUE

After modification, the dimensions are the same and all values in the matrix match the file in TMExplorer. If this was a new dataset, the modified geo object would be what you submit as the genes x cells matrix.

This dataset also has cell type information available. In order to prepare that for TMExplorer we’ll need to load in the file and remove the extra columns.

geo_cellinfo <- read.csv('GSE75688_final_sample_information.txt',sep='\t')
head(geo_cellinfo)

##    sample type index index2 index3
## 1 BC01_02   SC Tumor  Tumor  Tumor
## 2 BC01_03   SC Tumor  Tumor  Tumor
## 3 BC01_04   SC Tumor  Tumor  Tumor
## 4 BC01_05   SC Tumor  Tumor  Tumor
## 5 BC01_06   SC Tumor  Tumor  Tumor
## 6 BC01_08   SC Tumor  Tumor  Tumor

geo_cellinfo$type <- NULL
geo_cellinfo$index <- NULL
geo_cellinfo$index2 <- NULL
names(geo_cellinfo)[2] <- "label"
dim(geo_cellinfo)

## [1] 528   2

head(geo_cellinfo)

##    sample label
## 1 BC01_02 Tumor
## 2 BC01_03 Tumor
## 3 BC01_04 Tumor
## 4 BC01_05 Tumor
## 5 BC01_06 Tumor
## 6 BC01_08 Tumor

This file contains the cell type for each cell in the original dataset, as shown below.

table(geo_cellinfo$label)

## 
##   Bcell  Immune Myeloid Stromal   Tcell   Tumor 
##      83       4      38      23      54     326

shooshtarilab/TMExplorer documentation built on June 7, 2022, 1:31 p.m.