T cell state identifier (TCellSI) is a tool to access eight distinct T cell states including Quiescence, Regulating, Proliferation, Helper, Cytotoxicity, Progenitor exhaustion, Terminal exhaustion, and Senescence. TCellSI provides T cell state scores (TCSS) for samples using specific marker gene sets and a compiled reference spectrum of T cell states from transcriptomic data. The major algorithm of TCellSI is shown as follows:
You can install the development version of TCellSI by:
# install.packages("devtools")
devtools::install_github("GuoBioinfoLab/TCellSI")
library(TCellSI)Sample_expression: Complete gene expression data.frame in TPM format by log2-transformed RNA-seq data.
sample_expression <- TCellSI::exampleSample
# sample_expression
SRR5088825 SRR5088828 SRR5088830 ...
5_8S_rRNA 0.000000 0.000000 0.0000000
5S_rRNA 0.000000 0.000000 0.4346853
7SK 0.0c0000 0.000000 0.0000000
A1BG 1.907599 3.284418 4.0821248
A1BG-AS1 3.083914 2.021501 4.5169002
...
ResultScores <- TCellSI::TCSS_Calculate(sample_expression, ref = TRUE) (Optionally, ref = FALSE)
Output: The output of the function is a data.frame with TCSS metrics, where each row corresponds to a T cell state and each column represents a sample name.
# ResultScores
SRR5088825 SRR5088828 SRR5088830
Quiescence 0.6422158 0.6153140 0.5105856
Regulating 0.4404173 0.3974391 0.1670356
Proliferation 0.5780474 0.5761891 0.3994404
Helper 0.5647604 0.4063729 0.2121070
Cytotoxicity 0.6482720 0.5826073 0.1829111
Progenitor_exhaustion 0.5905679 0.5308304 0.2482848
Terminal_exhaustion 0.6624943 0.6222188 0.3311532
Senescence 0.5921543 0.5725185 0.2896508
If you want to apply this method to other states that interest you, you should compile a reference spectrum and prepare specific marker gene sets of your states. You can then calculate the scores for your states using the following function. If you choose not to provide a reference expression spectrum but to do the calculation directly, you can use the parameter ref=FALSE and do not need to provide the reference parameter.
OtherScores <- TCellSI::CSS_Calculate(sample_expression, ref=TRUE, reference = XXX, markers = XXX)
Forms of reference and markers look like:
#reference
#The self-constructed reference should contain log2-transformed, TPM-normalized gene expression data from RNA-seq or scRNA-seq.
cell_state1 cell_state2 cell_state3 ...
DDX11L1 0.32323232 0.54567463 0.32456323
WASH7P 0.82670591 1.89565638 1.40492732
MIR6859-1 0.02172025 0.03816506 0.52313432
MIR1302-2HG 0.00000000 0.00000000 0.00032302
MIR1302-2 0.00000000 0.00000000 0.00002132
...
#markers: A list of multiple cell states containing specific gene sets
#The number of marker genes per cell state can vary.
$cell_state1
[1] "XXX" "XXX" "XXX" ...
$cell_state2
[1] "XXX" "XXX" "XXX" ...
$cell_state3
[1] "XXX" "XXX" "XXX" ...
TCellSI still shows excellent results in the calculation of single-cell data and can assist in single-cell annotation.
In terms of operation. First, you should extract the count expression of single-cell data by reading the count file of single-cell data directly or seurat_obj@assays$RNA@counts in the seurat object. We suggested to convert it to the log(TPM +1) format. Then you can use TCellSI to perform calculations of the states scores for each cell of the single-cell data.
scRNA_scores <- TCellSI::TCSS_scRNAseqCalculate(sample_scRNA, core= XXX, ref = TRUE) # core: default value is 4; Optionally, ref = FALSE
Then you can add the score value of the result of the calculation into the metadata data box of the seurat object.
FeaturePlot(object = seurat_object, features = "TCSS") #viewing the distribution of scores in a umap; "TCSS" is the name of the column in which the categorical value is added to the metadata object
In addition, if you have an single-cell population annotation, you can create pseudobulk samples and then calculate the state scores for each samples, which can reduce the problem of drop-out in the single-cell data that leads to less accurate results. The creation of the pseudobulk is as follows:
How to create pseudobulk samples from single cell data ? If you want to do this, you should prepare an expression data. In this data, each row represents a gene and each column represents a cell ID (see example as follows). Also, you should prepare a single-cell annotation file, which includes columns of cell annotation and cell ID in expression file (see example as follows).
# expression data
NP710.20180123 NP711.20180123 NP71.20180123 ...
A1BG 0.070079488 0.216835131 6.269805313
NAT2 0.002001509 0.003654851 0.003190016
ADA 0.085464008 0.088970085 0.057264107
...
# single-cell annotation file
UniqueCell_ID annotation
NTH5.20180123 CD4_C01_CCR7
NTH64.20180123 CD4_C01_CCR7
NTR57.20180123 CD4_C01_CCR7
...
Then, you can use the following function to get pseudobulk samples.
pseudo_bulk <- TCellSI::create_pseudo_bulk(
annotation_data = XXX,
expression_data = XXX,
cluster_col = "annotation", # the column names of annotation in single-cell annotation file
cell_id_col = "UniqueCell_ID", # the column names of Cell_ID in single-cell annotation file
n_clusters = 18, # number of cell types annotated
factor = 5, # number of samples for downsampling, default is 5
sampling_rate = 0.6 # percentage of cells downsampled, default is 0.6
)
# see examples of the result
# pseudo_bulk, each column represents a newly pseudobulk samples, each row represents a gene.
CD4_C01_CCR7_bulk CD4_C01_CCR7_bulk.1 CD4_C01_CCR7_bulk.2
A1BG 0.495165739 0.67542360 0.737122107
NAT2 0.006033183 0.00337272 0.007104438
ADA 0.855647562 1.06058830 0.898952625
Result <- TCSS_Calculate(pseudo_bulk)
Please cite our paper if TCellSI is helpful.
- Yang, Jing‐Min, NanZhang, Tao Luo, Mei Yang, Wen‐Kang Shen,Zhen‐Lin Tan, Yun Xia, et al. 2024. "TCellSI: A Novel Method for T Cell State Assessment and Its Applications in Immune Environment Prediction."iMeta e231. https://doi.org/10.1002/imt2.231
