scRank

Cells respond divergently to drugs due to the heterogeneity among cell populations,thus it is crucial to identify the drug-responsive cell population for accurately elucidating the mechanism of drug action, which is a great challenge yet. Here, we address it with scRank using a target-perturbed gene regulatory network (tpGRN) to rank and infer drug-responsive cell population towards in-silico drug perturbation for single-cell transcriptomic data under disease condition. scRank enables the inference of drug-responsive cell types for single-cell data under disease condition, providing new insights into the mechanism of drug action.

Installation

• install dependent packages devtools and rTensor

#install.packages("devtools")
#devtools::install_github("rikenbit/rTensor")
devtools::install_github("ZJUFanLab/scRank")

Note

Key Updates

• Disease Relevance and Drug Effects Analysis: Introducing the new scRank_GSEA() and plot_drug_function() functions for analyzing disease relevance and drug effects in the highest-ranking cell types.
• Drug Type Specification: Added a type parameter in rank_celltype() to specify modeling effects of either agonists or antagonists, enhancing the versatility of drug response modeling.
• Efficient Large Matrix Manipulations: Integration of the Python module "tensorly" in the Constr_net() function, with a new parameter use_py, to optimize large-scale data processing.
• Enhanced Cell State Discernment: Integration of the scSHC algorithm into the CreateScRank() function with an if_cluster parameter, improving the tool's ability to discern various cell states. More about scSHC.
• Incorporating Drug Resistance Mechanisms: The resistance_target parameter in rank_celltype() allows for inputting targets of alternative pathways, aiding in the consideration of drug resistance mechanisms.
• Flexible Edge Weight Adjustment: Introduction of the keep_ratio parameter to adjust edge weights in the gene regulatory network, allowing for differential treatment of node types.

To-Do

• Packaging and Accessibility: We are in the process of submitting scRank to Bioconductor or CRAN for enhanced accessibility.

Overview

scRank method consists of two components, wherein the first is to reconstruct the gene regulatory network from expression ptrofiles using Constr_net function and the second step is to estimate the extent of the in silico drug perturbation for GRNs in each cell type using rank_celltype function.

scRank start with create a S4 object by CreateScRank function:

• the input is the gene expression profil eand meta is the cell type information.
• cell_type is the column name of the cell type information in meta
• species is the species of the data. ("mouse" or "human")
• drug is the drug name and target is the target gene of the drug. drug could be found in our database utile_database. if you know the specific target gene of the drug, you can input the target gene into target without inputing drug.
• type characters meaning the MOAs of drug including antagonist or agonist. Default is antagonist.
• if_cluster A logical meaning whether clustering single-cell transcriptomic data. Default is FALSE.

CreateScRank <- function(input,
                         meta,
                         cell_type,
                         species,
                         drug,
                         target,
                         type,
                         if_cluster,
                         var.genes)

The format of the input is as follows:

1. gene expression profile formatted by matrix or data frame, where the column is gene and the row is cell.
2. Seurat object with metadata containing cell type information

The meta is required if input is not a Seurat objectas, where its format as follows:

1. a dataframe with row names as cell names matched with column names of input and column names as cell type information cooresponding to the cell_type argument.

Tutorial

In this tutorial, we will demonstrate how to infer the drug-responsive cell type by scRank based on a demo dataset (GSE110894) containing BET inhibitor resistant and sensitive leukaemic cells.

1. Load the data and create a scRank object

we load the demo dataset from Seurat object, the drug target is known as Brd4.

seuratObj <- system.file("extdata", "AML_object.rda", package="scRank")
load(seuratObj)
obj <- CreateScRank(input = seuratObj,
                    species = 'mouse',
                    cell_type = 'labels',
                    target = 'Brd4')

2. Construct the gene regulatory network

obj <- Constr_net(obj)

3. Rank the cell types

obj <- rank_celltype(obj)

the final infered rank of cell types that determine the drug response is stored in obj@cell_type_rank

4. Visualize the result

To visualize the rank of cell types in dimension reduction space, we can use the plot_dim function.

plot_dim(obj)

To visualize the modularized drug-target-gene subnetwork in a specific cell type, start by initializing the modularization using the init_mod function. Afterward, you can use the plot_net function to display the network. The mode parameter in plot_net allows you to choose between a "heatmap" or a "network" for different types of visualization.

obj <- init_mod(obj)

plot_net(obj, mode = "heatmap", celltype = "sensitive")
plot_net(obj, mode = "heatmap", celltype = "resistant")

Reference

Chengyu Li, et al. scRank infers drug-responsive cell types from untreated scRNA-seq data using a target-perturbed gene regulatory network, Cell Reports Medicine, 2024, https://doi.org/10.1016/j.xcrm.2024.101568