TILPRED is an R Package for the classification of tumor-infiltrating T lymphocytes (TILs) from single-cell RNA-seq data.
TILPRED is no longer maintained. We recommend to use instead ProjecTILs
TILPRED is a logistic regression-based classifier that reads a SingleCellExperiment object containing CD8 T cell profiles and assigns to each cell a probability score of belonging to any of the following reference CD8 TIL transcriptomic states:
- Exhausted: a.k.a terminally exhausted cells (Tex), associated with terminal differentiation in the context of sustained antigenic stimulation. Phenotypically characterized by co-expression of inhibitory receptors (eg PD-1(Pdcd1), TIM3(Havcr2)), transcription factor Tox, cytotoxic molecules (e.g. Gzmb) and lack of Tcf1 (Tcf7) expression
- MemoryLike: a.k.a progenitor of exhausted cells (Tpex), also associated to sustained antigenic stimulation but retain capacity to self-renew and give rise to exhausted cells. Phenotypically characterized by co-expression of Tox, PD-1 (Pdcd1), and Tcf1 (Tcf7)
- EffectorMemory: antigen experienced T cells with effector memory features (e.g. co-expression of cytotoxicity genes such as Gzmk and Gzmb, and memory genes such as Tcf7, Lef1, Il7r and Ly6c2). These cells have low to intermediate expression of PD-1, and resemble CD8 T cells found upon acute infection (i.e. in the absence of sustained antigenic stimulation)
- Naive: Naive-like CD8 T cells (high expression of Tcf7, Lef1, Il7r, no expression of cytotoxicity genes or T cell activation markers such as CD44, CD69, etc.)
In addition, it predicts proliferation/cycling in each cell, independently of the CD8 TIL subtype. TILPRED uses gene rankings only and therefore is robust to different data normalization strategies. It was tested with scRNA-seq data produced with plate-based (smart-seq2) and droplet-based (10X 5' and 3' counting) technologies.
Before computing CD8 T cell states probabilities, TILPRED will automatically detect non CD8 T cell types. Non CD8 T cells are classified based on curated gene signature enrichment into: Treg (Foxp3 Regulatory T cells), CD4T (non Treg CD4+ T cells), NKT (NK T cells), Tcell_unknown (T cells of other kinds) and Non-Tcell (for cell types other than T cells, e.g. Myeloid, B cells, NKs)
Details on the reference CD8 TIL transcriptomic states and TILPRED construction and benchmarking are available in Carmona SJ et al. 2020
NB: TILPRED classifies CD8 TILs from mouse only. TILPRED using parameter human=T will only discriminate human T cells from non-T cells
TILPRED requires doParallel, doRNG and the Bioconductor packages AUCell and SingleCellExperiment
install.packages(c("doParallel","doRNG"))
if (!requireNamespace("BiocManager")) install.packages("BiocManager")
BiocManager::install(c("AUCell","SingleCellExperiment"))
library("SingleCellExperiment")
library("AUCell")
To install TILPRED directly from the Git Repo use remotes
if (!requireNamespace("remotes")) install.packages("remotes")
remotes::install_github("carmonalab/TILPRED")
library(TILPRED)
Run TILPRED on a SingleCellExperiment object containing the single-cell expression matrix of CD8 T cells
data(B16CD8TIL_SCE) # example SingleCellExperiment object
sce.pred <- predictTilState(data=B16CD8TIL_SCE)
View output
table(sce.pred$predictedState)
head(sce.pred$stateProbabilityMatrix)
For a running example please see this R Notebook
Santiago J. Carmona, Imran Siddiqui, Mariia Bilous, Werner Held & David Gfeller (2020) Deciphering the transcriptomic landscape of tumor-infiltrating CD8 lymphocytes in B16 melanoma tumors with single-cell RNA-Seq, OncoImmunology, 9:1, DOI: 10.1080/2162402X.2020.1737369