A Brief Note on Single-Cell RNA-Seq Analysis

4 minute read

Published:

A brief note highlighting the workflow of single-cell RNA-seq data analysis.

Contents

1. Pre-Processing

Goal:

  • To pre-process the scRNA-seq count matrices so as to obtain different levels of pre-processed data for different downstream analyses

1.1 Quality Control

Goal:

  • To remove outlier barcode data produced by dying cells, broken cells, doublets, etc.

Covariates used for thresholding:

  • The number of counts per barcode (count depth)
  • The number of genes per barcode
  • The fraction of counts from mitochondrial genes per barcode

Notes:

  • Consider all covariates jointly
  • Revisit QC if downstream analysis is not satisfactory
  • Adjust QC thresholds for different samples if necessary
[back to top]

1.2 Normalization

Goal:

  • To remove unwanted technical variation coming from count sampling

Notes:

  • It’s more common to normalize the count data by cells than by genes
  • Normalized data should be log(x+1)-transformed for downstream analysis that assumes data are normally distributed
  • Quality-controlled or normalized data can also be used for statistical testing of gene expression
[back to top]

1.3 Data Correction

Goal:

  • To further remove unwanted technical and biological variation

Source of variation:

  • Technical effects
    • Count depth
    • Batch
      • Between cell groups in an experiment (solution: batch correction)
      • Between experiments in a laboratory (solution: data integration)
      • Between datasets from different laboratories (solution: data integration)
    • Dropout (solution: imputation/denoising/expression recovery)
  • Biological effects
    • Cell cycle
    • Mitochondrial gene expression

Notes:

  • Corrected data can also be used for visual comparison of gene expression
[back to top]

1.4 Dimensionality Reduction

Goal:

  • To summarize and visualize the data in order to make downstream analysis computationally easier and more intuitive

Notes:

  • Typically 1,000 to 5,000 highly variable genes are firstly selected via feature selection
  • The feature number is further reduced by dedicated dimensionality reduction algorithms
  • Dimensionality reduction methods should be considered separately for summarization and visualization
[back to top]

2. Downstream Analysis

Goal:

  • To extract biological insights and describe the underlying biological system

2.1 Cluster Analysis

Goal:

  • To explain the heterogeneity in the data based on a categorization of cells into groups

Methods:

  • Clustering: to organize cells into clusters
  • Compositional analysis: to analyse clustered data in terms of the proportions of cells that fall into each cell-identity cluster
  • Cluster annotation: to find marker genes to characterize each cluster and annotate it with a meaningful biological label
[back to top]

2.2 Trajectory Analysis

Goal:

  • To regard the data as a snapshot of a dynamic process and investigate this underlying process
  • By representing the clusters as nodes and the in-between trajectories as edges, one can represent both the static and dynamic nature of the data

Methods:

  • Trajectory inference: to capture transitions between cell identities, branching differentiation processes, or gradual, unsynchronized changes in biological function
  • Metastable states identification: to investigate cellular densities along a trajectory, and find dense regions which may represent metastable transcriptomic states
  • Gene expression dynamics: to model the genes that vary smoothly across pseudotime in order to characterize the trajectory and identify the underlying biological process
[back to top]

2.3 Gene-Level Analysis

Goal:

  • Rather than describing the cellular heterogeneity, gene-level analysis uses this heterogeneity as context in which gene expression is to be understood

Methods:

  • Differential expression analysis: to investigate whether any genes are differentially expressed under different experimental conditions
  • Gene-set (pathway) analysis: to facilitate the interpretation of long candidate gene lists by grouping the genes into sets based on shared characteristics and testing whether these characteristics are overrepresented in the list
  • Gene regulatory network inference: to uncover the regulatory interactions among different genes and small molecules
[back to top]

References

  1. Current best practices in single-cell RNA-seq analysis: a tutorial. MD Luecken, FJ Theis. Molecular systems biology, 2019
  2. Computational Methods for Single-Cell Data Analysis. GC Yuan. Springer, 2019
  3. Orchestrating single-cell analysis with Bioconductor. RA Amezquita, et al. Nature methods, 2019
  4. Single-cell RNA sequencing technologies and bioinformatics pipelines. B Hwang, JH Lee, D Bang. Experimental & molecular medicine, 2018
[back to top]

You May Also Enjoy