Skip to content

README.md

Latest commit

 

History

History
68 lines (46 loc) · 2.89 KB

README.md

File metadata and controls

68 lines (46 loc) · 2.89 KB
Raw

scTenifoldNet.lite

License: GPL (>=2)

The goal of scTenifoldNet.lite is to provide fast and memory-efficient implementation of the scTenifoldNet workflow for constructing and comparing single-cell gene regulatory networks. The computational performance for principal component regression, tensor decomposition, and manifold alignment was optimized, while preserving the core analytical framework of the original scTenifoldNet method.

Installation

You can install the development version of scTenifoldNet.lite like so:

devtools::install_github("ycli1995/scTenifoldNet.lite")

Package Overview

The original six-step pipeline in scTenifoldNet has been optimized for memory and speed. Below describes the main differences.

scQC: Quality control

This step filters cells by library size, outlier detection, minimum gene expression fraction, and mitochondrial read ratio.

  • I add drop = FALSE for subsetting to keep the matrix formats.

cpmNormalization: Counts-per-million (CPM) normalization

  • Use RcppEigen to speed up.

makeNetworks: Constructs gene regulatory networks from subsampled cells

  • The GRNs are constructed using principal component regression (pcNet).
  • I add a fast = TRUE parameter to use global principal component scores among all genes, which avoids running RSpectra::svds() upon each gene to speed up.

tensorDecomposition: Network denoising

  • I only keep the 3D CANDECOMP/PARAFAC (CP) tensor decomposition path.
  • The denoised tensor will be represented as U_list instead of a memeory-expensive array.

manifoldAlignment: Non-linear manifold alignment

  • The W matrix with 2n * 2n elements will not be constructed to save memeory.

dRegulation: Differential regulation testing

  • This step testing the differential regulation on the manifold alignment output.
  • No need to optimize.

Example

The motivation of this package is to perform scTenifoldKnk() pipeline for virtually knockout:

library(scTenifoldNet.lite)

scTenifoldKnk(countMatrix = scRNAseq, gKO = "G100", qc_minLSize = 0, fast = TRUE)

Citation

If you use scTenifoldNet.lite in your research, please cite the originial manuscript:

Osorio, D., Zhong, Y., Li, G., Huang, J. Z., & Cai, J. J. (2020). scTenifoldNet: A Machine Learning Workflow for Constructing and Comparing Transcriptome-wide Gene Regulatory Networks from Single-Cell Data. Patterns, 1(9), 100139. doi:10.1016/j.patter.2020.100139

Osorio, Daniel, et al. "scTenifoldKnk: An efficient virtual knockout tool for gene function predictions via single-cell gene regulatory network perturbation." Patterns 3.3 (2022).