CosMx

📝 Description

Scripts for the analysis of Nanostring CosMx data.

🎯 Aim

Perform QC, pre-processing, clustering and visualisation of CosMx data.

Getting data from AtoMx

• Export Seurat object, tiledb object, flat files, and raw data from AtoMx

🔍 Details

R Scripts:

• These scripts use the tileDB output from AtoMx.
• Load tileDB data object the R environment.
• Raw counts, normalised counts and meta data can all be extracted from the tileDB object.
• CosMx_QC.R
  • QC metrics are visualised and saved to files
• CosMx_PreProcess.R
  • UMAP and initial clusters calculated by AtoMx are extracted and visualised.
  • As the data is not filtered the initial UMAP and clustering results are not accurate.
  • For instance, an excessive number of clusters are identified. We note that the number of clusters is reduced when these cells are removed.
  • Cells are visualised in their spatial context - in 2D space on the slide.
  • Cells can be visualised by QC flag or cluster ID.
  • Cells are scored for cell type signatures - in order to identify hybrid cells
  • All metadata is saved and seurat objects saved as RDS files
• CosMx_PostProcess.R
  • RDS files saved in previous script are loaded
  • Add hybrid status to the seurat object meta data nd then remove hybrid cells
  • Re-process data - find variable features, scale, normalise, cluster etc...
  • Use scType for automated cell type annotation
  • Compare EGFR+ vs EGFR- tumour cells

squipy scripts

• See squidpy docs: https://squidpy.readthedocs.io/en/stable/notebooks/tutorials/tutorial_nanostring.html
• This script uses raw data exported from AtoMx to perform analysis in jupyter notebook
• Raw data exported from AtoMx needs to be treated according to documentation here:
  • https://nanostring-biostats.github.io/CosMx-Analysis-Scratch-Space/posts/squidpy-essentials/squidpy-essentials.html#sec-noimage
  • In a single folder for each sample we need the following (copy files and folders as required, see CosMx_Data_Prep.sh for functions needed to copy over files - note this is a guide not a script.):
    • Flat files
    • CellComposite folder
      • Found in RawFiles///CellStatsDir
    • CellLabels folder
      • CellLabels files are in individual FOV folders in the CellStatsDir - need to copy them all into a folder
    • CellOverlay folder
      • Found in RawFiles///CellStatsDir
    • CompartmentLabels folder
      • CellLabels files are in individual FOV folders in the CellStatsDir - need to copy them all into a folder
• Once data is in a single folder in the correct folder format the squidpy script can be used
• Launch an OnDemand session on Apocrita using jupyter notebooks - a lot of memory is required - e..g. 2 cores, 128GB memory
• Install relevant packages needed and then run the notebook