Skip to content

scRNAseq Analysis Recipe

Stephan Reichl edited this page Nov 12, 2024 · 5 revisions

The scRNA-seq Analysis Recipe takes you from raw count matrices to enrichment analysis results of your differentially expressed genes while providing unsupervised analyses and genome browser tracks for visualization and quality control.

flowchart LR;
    ngs_fetch-->scrnaseq_processing_seurat;
    scrnaseq_processing_seurat-->genome_tracks;
    scrnaseq_processing_seurat-->unsupervised_analysis;
    scrnaseq_processing_seurat-->dea_seurat;
    unsupervised_analysis-->dea_seurat;
    dea_seurat-->enrichment_analysis;
Loading

Modules

The following Modules are used in this Recipe:

  1. (optional) Fetch publicly available bulk ATAC-seq data (coming soon).
  2. scRNA-seq Data Processing & Visualization for processing and preparing a Seurat object for all downstream analyses.
  3. Genome Browser Track Visualization for quality control and visual analysis of genomic regions of interest or top hits.
  4. Unsupervised Analysis to understand and visualize similarities and variations across cells, including dimensionality reduction and cluster analysis.
  5. Differential Analysis using Seurat to identify and visualize statistically significantly differentially expressed genes between groups.
  6. Enrichment Analysis for biomedical interpretation of differential analysis results using prior knowledge.
  • add pseudobulk analysis as orthogonal analysis that adds modeling capabilities

Strategy

Cell type assignment strategy

  • Gene expression and marker signatures (UMAP, violin-, dot plots).
  • Dimensionality reduction and clustering results, distance ~ similarity (UMAP). In combination with clustree results other resultions, i.e., splitting the data in more clusters, could be warranted.
  • DEGs between clusters, especially up-regulated compared to all others are good indicators of cell type.
  • preranked GSEA of DEG profiles, especially using databases that relate to your data are useful e.g., Allen Brain Atlas for neurscience.
  • GPTCelltype: AI models from OpenAI (e.g., GPT-4o) ore predict cell types based on genes per cluster (e.g., top most significant up-regulated genes) and tissue of origin (optional). The o1-preview model is claimed to be the most comprehensive one and performed well in our hands, but we recommend running multiple models and use consistency between their predictions as confidence measure.

--- MORE COMING SOON ---

Templates for a Methods section of a scientific publication can be found in each Module's README.

Data

--- COMING SOON ---

Code & Configuration

--- COMING SOON ---

Results

--- COMING SOON ---