Analysis completed on behalf of the MGH KFCCR Tumor Cartography Core for the Ellisen Lab.
- Steps 0-8 require that all code is run in a micromamba environment made from the YML file provided in order to reproduce results exactly.
- Steps 9-10 were completed outside of this environment. Plots from Step 9 specifically will be near impossible to reproduce exactly, since I forgot to set the seed when I made the plots initially and sent them over for the paper.
All raw and/or processed data can be requested by emailing the corresponding author.
Setting up the unprocessed GeoMx dataset object.
Performing quality control and aggregating the counts among sets of probes that correspond to the same genes.
Q3-normalizing the data for downstream use with other NanoString tools, such as SpatialDecon, as well as exploring some alternative normalization methods.
Ensuring that the three tissue compartments (tumor, fibroblast, and immune) have marker genes that make sense as a sanity check.
Performing differential expression analysis with limma and exploring other methods, such as mixed effects modeling with lme4.
Deconvolving the immune AOIs to assess abundances of immune cell types, using SpatialDecon.
Deconvolving the fibroblast AOIs to assess abundances of CAF cell types, using SpatialDecon.
Exploring some possible improvements in the gene selection and normalization processes for making PCA plots.
Exploring the GeoDiff package, which offers an alternative QC and normalization workflow for this kind of data.
Performing gene set enrichment analysis, using the limma results from Step 4. Publication figures:
- TROP2 paper Figure S1C (as noted above, this figure may be difficult to reproduce exactly, due to failure to set the seed)
Creating some specific plots and writing up computational methods for publication. Publication figures:
- TROP2 paper Figure 1C and 1E
- TROP2 paper Figure S1B