Supplementary code for paper
Each script is named after an associated figure in the paper. While each script may be run with the appropriate interpreter (bash for .sh files, Rscript for .R files, python for .py files, stack for .hs files), input files and folders must be changed to represent the location of the raw data on the user's local system. Many of these scripts use too-many-cells, which can be download at https://github.com/GregorySchwartz/too-many-cells with instructions and documentation for additional features.
scRNA-seq clustering, doublet removal, cell type classification, and doublet identification using DoubletFinder
Seurat v3.1.5 was used for filtering, UMAP generation, and initial clustering. Please refer to the FigS1-S3_Seurat_and_DoubletFinder.Rmd for relevant code to be run in R.
In addition to DoubletFinder (described above), Scrublet v0.2.1 was also used to identify doublets. Please refer to FigS2A_Scrublet.py script for relevant code to be run using Python.
Garnett was used for cell classification. Please refer to FigS3_Fig1_Garnett.Rmd for relevant code to be run in R. The cell type marker file used in conjunction with the code is found in supplemental materials associated with the paper.
The label transfer script is a small command line program to use Seurat's label transfer on single-cell data. Usage is:
Rscript Fig_S4BCEF_label_transfer.R LABELPATH OUTPUTPATH LOGNORMFLAG FILTERANCHORS INPUT REFINPUTS
Here, LABELPATH are the labels for each barcode in a file (see too-many-cells documentation), OUTPUT is the output label, LOGNORMFLAG is whether to use Seurat's LogNormalize instead of SCT, FILTERANCHORS is the number of anchors to use, the INPUT is the input scRNA-seq data in matrix market or CSV format, and REFINPUTS are a list of reference scRNA-seq data matrices. get_mat.R in this repository collects helper functions for this script.
stack Fig_5AF_all_IMC_CyTOF_trees_analyze_norm.hs, (installing stack: https://docs.haskellstack.org/en/stable/README/) will generate trees used in the paper. On line 60, replacing any part of the tuple will cahnge the arguments to too-many-cells called. In order, the tuple specifies (see too-many-cells documentation): (tree cutting number, a name for the analysis, a path for the input scRNA-seq matrices, a whitelist of included cells, the normalization used to process the data, how many dimensions to drop with LSA, whether to specify a specific node to set as a new root, the labels file containing labels per cell barcode).
submitted_overlayDiscreteLabels.sh will overlay the cell labels/annotations (specified using --labels-file) on the too-many-cells tree (specified using --prior) colored by cell type (specified using --draw-colors). The tree can also be colored by protein expression using the --draw-leaf option as done in the script submitted_dualMarkerOverlay.sh (e.g. Figure 5A, 5F in the manuscript) or by the labels indicating which cells are close or nearby a group of cells from a node (specified using -j or --projection-file) as done in submitted_overlyByProximityToBase.sh.
Analysis plots such as pie charts showing disease state composition in a cell type (e.g. Figure 5b, 5G), boxplot to check percentage of cell types in each disease state for every donor (e.g. Figure 5C, 5H), and other supplementary plots can be generated using the script submitted_quantification.py in IMC/CyTOF folder.
Differential genes were found using edgeR through TooManyCells with the -–normalization "NoneNorm" to invoke edgeR single-cell preprocessing, including normalization and filtering.
too-many-cells differential --prior "" --matrix-path "TotalMed Normalized Matrix" --labels-file "" --normalization "NoneNorm" --nodes "(,)" --labels "([\"\"], [\"\"])" -t 1000000 +RTS -N35
For cell labels relevant to the scRNA-seq data, please refer to the file located here: https://www.dropbox.com/s/1i065hbai0vyfue/fasolino_et_al.rds?dl=0
Differential genes were found using edgeR through muscat. Please refer to DE_PseudoBulk_analysis_edgeR_Table11_12_13.R for relevant code to be run in R.
Pseudobulk of samples were aggregated using muscat. Correlation analysis of each individual gene with GAD levels were calculated. A p-value of less than 0.05 and correlation greater than 0.99 was used to filter. Please refer to Correlation_PseudoBulk_AAB_Fig3A_3B.R for the relevant code to be run in R.
The different expression specificity metrics were calculated using CELLEX tool. The output of CELLEX and GWAS traits were given as input to CELLECT which ranked association of cell types and various disease studies based on p-value. Please refer to cellex_prior_plot_Fig_S8D.R to generate the plot in R.
Compare average expression of top 3 DEGs across all cell types in T1D vs. Control, T1D vs. AAB, and AAB vs. Control. Please refer to top3_geneex.R for relevant code to be run in R.