Analysis

Analysis

Code by Leo

Summary

• convert_sce.R: builds the SingleCellExperiment Bioconductor object and defines some plotting functions used in other scripts.
• sce_scran.R: determines the 1942 variable genes, computes PCs and saves the sce object (file Human_DLPFC_Visium_processedData_sce_scran.Rdata) used by other scripts. Also explores several things you can do based on https://osca.bioconductor.org/ and the scran vignette.
• sce_zinbwave.R: uses the sce object from sce_scran.R and runs zinbwave.
• sce_image.R: uses the sce object from sce_scran.R and runs several k-means approaches using the clusteR package.

convert_sce.R

Related files:

• /dcs04/lieber/lcolladotor/with10x_LIBD001/HumanPilot/Analysis/Human_DLPFC_Visium_processedData_sce.Rdata
• /dcs04/lieber/lcolladotor/with10x_LIBD001/HumanPilot/Analysis/geom_spatial.Rdata

Details:

• Builds the sce object with image data under metadata(sce)$image which is a single data.frame. Subsetting doesn't automatically subset the image, so you have to do it yourself when plotting.
• Creates some of the functions and the ggplot2 geom_spatial() layer used for plotting later on. Newer and more complex versions are in the global.R file from spatialLIBD. In particular, the colors are different between these versions as now spatialLIBD uses the original colors from the 10X scripts when the number of clusters <= 12 and then relies on Polychrome::palette36.colors() (which would break if more than 36 colors are needed).

sce_scran.R

Related files:

• pdf_scran
• rda_scran
• /dcs04/lieber/lcolladotor/with10x_LIBD001/HumanPilot/Analysis/Human_DLPFC_Visium_processedData_sce_scran.Rdata

Details:

• Defines the sce and top.hvgs objects used in other scripts and in spatialLIBD. As we decide to add more stuff to the sce object, I'll edit this R code as well.
• Builds a SNN graph with 50 nearest neighbors (K=50) that results in 28 clusters across all 12 images.
• Cuts that SNN graph to k = 4 up to 28.
• Mostly follows https://osca.bioconductor.org/.

It's long enough now that I'll continue new analyses in other scripts.

sce_zinbwave.R

Related files:

• pdf_zinbwave
• rda_zinbwave

Details:

• Runs zinbwave::zinbwave() and saves the results. Initially, it chose the genes using zinbwave but now it uses the same ones from sce_scran.R. It failed at running RSEC() either due to memory (when using kmeans) or because I stopped trying to get ClusteR::KMeans_rcpp to work with RSEC().

sce_image.R

Related files:

• pdf_image
• rda_image

Details:

• The name of the script comes from me thinking that we could process it like an image as in https://cran.rstudio.com/web/packages/ClusterR/vignettes/the_clusterR_package.html (the dog example). But then I realized that the image example didn't use X or Y information. Some of this quick exploration code is near the end of the script (just so I wouldn't lose the code forever).
• Ultimately, the likely useful output of this script is the set of k-means clusters from k = 4 up to 28 using ClusteR. Although I didn't save all the kmeans output (only the cluster labels), so we might need to re-compute it.
• This is also the script where I played the most with trying to add the X and Y information as well as a blocking factor (6 levels, one per each subject + slice (position) pair of images).