In ff98li/scRGNet: Perform scRNA-seq Analysis using Single-cell Graphical Neural Network (scGNN) Framework

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>",
  fig.path = "man/figures/README-",
  out.width = "100%"
)

scRGNet

Description

scRGNet is an R package for inferring cell-cell networks from encoded scRNA-seq data. It is the first R package that attempts to use torch[@torch] in R to implement an feature(that is, gene) autoencoder from the recently proposed single cell graph neural network (scGNN) framework.[@scGNN] It generates an encoded feature matrix containing the low-dimensional representation of gene expression in each cell, and build a Cell-Cell network from the feature matrix using KNN and isolation forest[@isolationForest][@isotree]. When using the feature autoencoder, discretized regulatory signals quantified from gene expression modeled by a left-truncated mixture Gaussian (LTMG) model can also be used as a regulariser.[@LTMG] It is unique from other R packages for scRNA-seq analysis in that scRGNet offers an option to analyse scRNA-seq data without assuming statistical distributions or relationships for gene expression.

• R requirement: 4.1.2 or later version
• Development environment: Ubuntu 20.04 LTS
• Development platform: x86_64-pc-linux-gnu (64-bit)

Installation

To install the latest version of scRGNet:

require("devtools")
devtools::install_github("ff98li/scRGNet", build_vignettes = TRUE)

To run the Shiny app:

scRGNet::runscRGNet()

Overview

library(scRGNet)
ls("package:scRGNet")
data(package = "scRGNet")

Note that there are two datasets included in this package gene_counts and gene_counts_small. gene_counts is a raw scRNA-seq matrix from experiment GSE138852[@GSE138852]. gene_counts_small is a subset of the gene_counts data for a quick demo of the package, containing only 48 cells and 1000 genes. For usage of functions in the package, please refer to package vignettes for more details:

browseVignettes(package = "scRGNet")

An overview of the package structure is provided below:

{ width=75% }

An overview of the package workflow is illustrated below:

Contributions

The author of the package is Feifei Li. The runLTMG function uses the LTMG object and the function for inferring LTMG tags from scgnnltmg[@scGNN]. data.table R package[@dt] is used for fast reading in a large size scRNA-seq raw matrix from csv. The Matrix[@matrix] R package is used to store scRNA-seq data as a sparse matrix to reduce memory usage, and used to convert a tensor object to an R matrix. The scDataset object is an R6 object[@r6] inherited from class dataset from torch. The feature autoencoder is also an R6 object inherited from the basic neural network modules nn_module from torch R package, and it makes use of its functional modules nnf_linear and nnf_relu.[@torch] Iteration of model training makes use of coro::loop form the coro R package. The model training also uses progress R package to inform users the model trainning progress. The generateNetwork function makes use of graph_from_data_frame from igraph R package to generate a plottable igraph object.[@igraph], and the isolation forest model from the isotree R package[@isotree] is used to prune outliers in cell graphs produced by generateNetwork. The interactive visualisation of produced cell network makes use of the visNetwork R package[@visNetwork]. plotDegree and plotLog make use of the graphics R package[@graphics]. cluster_label_prop and degree function from igraph R package are used to compute the communities and degrees of the network. The shiny app of this package is made with shiny[@shiny], shinyjs[@shinyjs], and shinybusy[@shinybusy] R packages. Except for the LTMG modeling in runLTMG uses external R package for computation, all other functions for data processing and analysis in this package are my original R implementation.

Acknowledgements

This package was developed as part of an assessment for 2021 BCB410H: Applied Bioinformatics, University of Toronto, Toronto, CANADA.

References

ff98li/scRGNet documentation built on Jan. 14, 2022, 4:58 a.m.