Decoding breast cancer: Unraveling subtype and model differences through multi-model single-cell RNA sequencing data integration

This repository contains a description and in-house scripts and code used to process the single cell data for Altman et.al manuscript.

Bioinformatics Pipeline Steps

Step 01: FastQC and MultiFastQC

Step	Details
Tool Name and Version	FastQC v0.11.9
Pipeline Script	01_fastqc.sh
Input File Format	Text file with one FastQ file name and absolut path per line.
Example Usage	~/01_fastqc.sh -f /path/to/fastqFiles.list -d /path/to/output/dir/

Step	Details
Tool Name and Version	MultiQC, version 1.17
Pipeline Script	None, manually run MultiQC in same directory as FastQC output.
Output Files used in downstream steps?	No, FastQC and MultiQC reports are manually reviewed to inform on data quality.

Step 02: Single Sample Alignment

Step	Details
Tool Name and Version	CellRanger, cellranger-6.0.1
Pipeline Script	02_runCellRanger.sh
Input File Format	No header, TAB delimited, has the following 5 columns: sample ID used in fastq file name, result directory name, reference genome (grch38, mm10, or grch38mm10), number of cells to target, absolute path to raw fastq data directory.
Example Usage	~/02_runCellRanger.sh -f /path/to/sampleFile.list -o /path/to/output/dir/ -n 35
Output Files used in downstream steps?	Yes: PDX Samples have a GEM file used to identify human cells in the step 3, removing mouse cells. Pure Human/Mouse or Organoid Samples utilize the filtered_bc_matrix data files as input into step 5, finding dead cells.
Output File formats and type	CellRanger “outs” directory with GEM file and "filtered_feature_bc_matrix" directory of files.

Step 03: Removal of Mouse Cells and Conversion Back to FastQ

Step	Details
Tool Name and Version	subset-bam 1.1.0
Tool Name and Version	bamtofastq v1.3.2
Pipeline Script	03_create_human_bam.sh
Input File Format	TAB delimited, has the following 3 columns: sample ID (same as Step 02), absolute path to samples “outs/analysis” directory, absolute path to samples “outs” directory.
Example Usage	~/03_create_human_bam.sh -f /path/to/sampleFile.list -d /path/to/output/dir/
Output Files used in downstream steps?	Yes, FastQ files used in step 4, realignment to human genome.
Output File formats and type	BAM and FastQ files that only contain human reads.

Step 04: Realign to Human only Genome

Step	Details
Tool Name and Version	CellRanger cellranger-6.0.1
Pipeline Script	02_runCellRanger.sh
Input File Format	See Step 02; must edit the number of cells to those that are human only instead of the default targeted number. Also add the "-F" for force cells option to stop CellRanger from doing further filtering of cells.
Example Usage	~/02_runCellRanger.sh -f /path/to/sampleFile.list -o /path/to/output/dir/ -n 35 -F
Output Files used in downstream steps?	Yes, Human only data: filtered_feature_bc_matrix data files are directly input into step 5, finding dead cells.
Output File formats and type	See Step 02

Step 05: Finding Dead Cells

Step	Details
Tool Name and Version	R 4.1.3
Tool Name and Version	hdf5-1.12.2
Pipeline Script	05_DeadCellAnalysis.R
Input File Format	TAB delimited, has the following 2 columns: sample ID, absolute path to human-aligned “outs” directory.
Example Usage (use --help option for all parameters)	Rscript 05_DeadCellAnalysis.R -r UniqRunID -c sample_tabdelim_file.list -f /path/to/mitogenes/MitoCodingGenes13_human.txt -s human -o /path/to/results/directory
Output Files used in downstream steps?	Yes, the CSV file “cells2keeps” listing the good quality barcodes is used in step 6, sample merging.
Output File formats and type	Tab delim .txt report, .png images of violin plots before and after filtering, and .csv “cells2keep” files listing barcodes to keep.

Step 06: Sample Merging

Step	Details
Tool Name and Version	R 4.1.3
Tool Name and Version	hdf5-1.12.2
Pipeline script	06_SeuratMerge.R
Input File Format	Comma delimited, has header row (SampleName,DataType,SamplePath,Source,Condition,Sex,TumorType,Cells2Keep), has the following 8 columns (but more can be added and will be used as cell annotations): Sample ID, data type (10X or seurat), absolute path to sample’s “filtered_feature_bc_matrix” directory, sample source (e.g. PDX, MGT, etc), condition or treatment, sample sex (M or F), tumor type (e.g. HCI011, UCD52, etc), absolute path to CSV file listing barcodes of cells to keep in merge.
Example Usage (use --help option for all parameters)	Rscript 06_SeuratMerge.R -r UniqRunID -c /path/to/input.csv -i LogNormalize -t simple --parallel -n 32 --filter
Output Files used in downstream steps?	Yes, .h5 file is used by CellRanger Reanalyze to generate a Loupe file
Output File formats and type	.h5 file, RData file contining the merged Seurat object, multiple .CSV annotation files (one for each column in the input CSV file)

Step 07: Create Loupe File

Note: During the development of this manuscript the RLoupe package that converts Seurat objects to Loupe-compatible files had not been released and this was our work-around.

Step	Details
Tool Name and Version	CellRanger cellranger-6.0.1
Pipeline script	N/A, calls the reanalyze function directly.
Input File Format	the .h5 file generated from the merging script in step 06.
Example Usage	cellranger reanalyze --id=UniquRunID --description=RunDescription --matrix=/path/to/merged_file.h5 --params=/path/to/reanalyze_params.txt
Output Files used in downstream steps?	Yes, Loupe file used for figure generation and analysis.
Output File formats and type	CellRanger “outs” directory with a Loupe File.

Once the Loupe file is generated, it is opened and the annotation CSV files from step 06 are manually imported. This includes the UMAP and tSNE layouts and the SNN clusters. Because this was a work-around and not a supported conversion method by 10X the layout and clustering of the default generated data in the Loupe file cannot be trusted as can be seen by the SampleID that ever only has 99 barcodes in it. The LoupeR package fixes most of this; however, at this point the annotations are being imported incorrectly.

Step 08: Identifying Malignant Cells using InferCNV

This step was only done once on a 33% subsampled dataset as using the entire dataset exceeded out compute resources.
This script only processes all samples in the input file in batches of "k" at a time. The same 5 manually selected samples are also run with each batch and are used to calculate the Zscores.

Usage NOTE: The MetaData CSV for inferCNV has to have 3 columns named exactly “Sample_ID,TissueType,inferCNV.normal”. The TissueType and inferCNV.normal columns are identical except the inferCNV column has "REFERENCE" in replacement of the tissue type for those samples that are to be used as the static number of reference samples to calculate the inferCNV scroe Z-score. Usage NOTE: The MetaData CSV for inferCNV has to use the EXACT name sample names that are used in step 06 for sample merging (i.e. the first column in the list CSV file).

Step	Details
Tool Names and Versions	R 4.1.3
Tool Name and Version	JAGS 4.3.1
Tool Name and Version	InferCNV 1.10.1
Pipeline Script	08_inferCNV_id_malignent_cells.R
Input File Format	The .RData file from the merging step.
Example Usage	Rscript 08_inferCNV_id_malignent_cells.R -r UniqRunID -o /path/to/existing/output/directory/ -i /path/to/merged_file.RData -k 5 -f /path/to/sample/phenotype_MetaData.csv -g /path/to/reference/genome/genes.gtf
Output Files used in downstream steps?	Yes, the CSV files were loaded into the Loupe File and used to identify clusters of cells that are normal-like.
Output File formats and type	CSV files with scores, RDA file with inferCNV object

The InferCNV scores were overlaid on the Loupe plot to identify if the normal samples correlated with low z-scores. This was done by discretizing the scores to round to the closest integer and then placing everything greater than 6 in a single group ">=6". Any non-normal cells that clustered with the normal reference groups were selected for removal to generate the Cancer-Only merge (step described below).

Step 09: Cancer-Only Merge

To generate the Cancer-Only datset, the 3 subsampled master merge Loupe files that included all samples was utilized to select the clusters in the UMAP plot that were deemed to be cancer cell clusters with a high z-score (i.e. clusters with normal samples and any non-normal cells that grouped with the normal samples were not selected and therefore not included in the Cancer-Only dataset). The chosen cancer-only barcodes were downloaded and concatenated into a single CSV file and input into the script "09_cells2keep_convertMerged2Unmerged.R", which divided them up into new "cells2keep" csv files for each sample and saved them in the original results directories.

Step	Details
Tool Names and Versions	R 4.1.3
Pipeline script	09_cells2keep_convertMerged2Unmerged.R
Input File Format	The ORIGINAL CSV file used to create the merged file from Step 06.
Input File Format	A list of cancer cell barcodes exported from the Loupe Browser. Must concatenate multiple barcode subsets into a SINGLE CSV file if the merge was done in groups.
Example Usage	Rscript 09_cells2keep_convertMerged2Unmerged.R -r UniqRunID -c /path/to/step06/input.csv -o /path/to/existing/output/directory/ -k /path/to/barcode/to/keep/list.tsv
Output Files used in downstream steps?	Yes, the CSV files are utilized by Step 06 to create the Cancer Only Merge.
Output File formats and type	CSV files, one file for each sample included in the master merge.

After creating the new cells2keep CSV files, the Cancer Only input CSV file needs to be generated. For this work we copied the file used in the first round of running Step 06 and deleted the Normal sample records. Then the "Cells2Keep" column was edited to point to the new CSV file that was created in Step 09. Step 06 was then run again to generate the Cancer Only merge, followed by Step 07 to generate the Loupe File.

Reference Step 06 and Step 07

Step 10: ER and TN Specific Merges

Two subsets of the Cancer Only dataset were created consisting of ER+ and TNBC samples only. The CSV inventory file utilized by Step 09 to create the cancer only merge was copied twice and all sample records subset to either the ER+ samples or TNBC samples. These were then utilized to create new merged datasets by running Step 06 and 07 again using these new library files.

Reference Step 06 and Step 07

Step 11: Pseudo PAM50 Subtyping

Step 12: SCSubtype Subtyping

Step 13: TNBCType Subtyping

Funding

[ADD FUNDING HERE]