Seurat Mito Prefix Removed from Features

Question

[Koki]

Hello,

I am looking for help on how to filter mitochondrial genes from a Seurat object. Running the following code results in a percent.mt of zero for all cells:

mito.genes <- grep(pattern = "^MT-", x = Features(seurat), value = TRUE, ignore.case = TRUE)

seurat[["percent.mt"]] <- PercentageFeatureSet(seurat, pattern = "^Mt-") # Any variation of MT, Mt, and mt
VlnPlot(seurat, features = c("nFeature_RNA", "nCount_RNA", "percent.mt"), ncol = 3)

Manually searching for mitochondrial genes using Features(seurat) shows mitochondrial genes are present in my features, but without the mt label. Thank you!

Answer 1

[LChart]

Atp8a1 is a nuclear-transcribed gene (chr4 in humans). Typically one is interested in removing mitochondrially-transcribed genes (chrM). Are you interested in genes relevant to mitochondrial biology, or only genes relevant to QCing live/highquality (nuclear abundant) vs dead (mitochondrially abundant) cells?

[Koki]

Thank you for the correction. I mistook Atp8a1 with MT-ATP8, but I am still interested in removing mitochondrial genes. I an analyzing GSM3017261 of the following dataset, which is provided in the form of a sparse matrix: https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE110823

Is it possible that the authors have filtered the mitochondrial genes prior to upload? Just want to make sure that mitochondrial genes are not being included by accident.

[LChart]

Very possible. The most sure way would be to use ensembl, look up the gene symbol to get the gene position, and remove anything with chrM.

Answer 2

[yura.grabovska]

If running:

mito.genes <- grep(pattern = "^MT-", x = Features(seurat), value = TRUE, ignore.case = TRUE)

Returns no elements in mito.genes then your dataset does indeed not have those genes in the list.

From the supplementary methods of the paper you cited:

Selecting high quality transcriptomes from the mouse CNS experiment

We discarded any transcriptomes with >1% reads mapping to mt-RNA, to ensure that all of our transcriptomes originated from nuclei. Transcriptomes with fewer than 250 expressed genes or greater than 2,500 expressed genes were also discarded. This resulted in retention of 163,069 transcriptomes. After clustering (see below), cells in putative doublet clusters were filtered as well, yielding 156,049 transcriptomes used for downstream analysis.

They specifically filtered for mitochondrial genes and this is intended to be sn-seq anyway

[Koki]

Thank you for the clarification! I guess I was confused on whether papers remove mitochondrial counts prior to uploading counts matrices or if I messed up my analysis. I've noticed some datasets keep the counts, such as in the Seurat vignettes, while others pre-filter.

[yura.grabovska]

There's no hard and fast rule because it's not a prerequisite of journals to have strict rules about single-cell data. Some authors provide cellranger outputs, some provide fastqs and some provide a sparse matrix or an .rds file with a scater/Seurat environment object. The key here really is to understand where your data is coming from and what it contains. I didn't do anything fancy, I just read the paper methods from the citation you provided and would always strongly encourage that practice.