Skip to content

CosMX SMI Starter Tutorial

Input Data

The input data is from an adult mouse hippocampus, extracted by masking a coronal brain section. The original full-section

File Format

The CosMx SMI by NanoString generates high-resolution spatial transcriptomics data with single-molecule resolution with a comma-separated values (CSV) table.

CSV File Format

1
2
3
4
"fov","cell_ID","x_global_px","y_global_px","x_local_px","y_local_px","z","target","CellComp"
64,0,-473043,7954.533,4015.3,4246.2,1,"Gfap","None"
64,0,-473022.9,7902.723,4035.48,4194.39,1,"Fth1","None"
64,0,-473132,7836.476,3926.34,4128.143,1,"Ptn","None"
  • fov: The field of view (FOV) number.
  • cell_ID: Unique identifier for a single cell within a given FOV. 0 if background or unassigned molecules.
  • x_global_px, y_global_px: Global pixel coordinates relative to the tisse.
  • x_local_px, y_local_px: The x or y position (in pixels) relative to the given FOV.
  • z: Z-plane index representing the depth (optical section) where the transcript was detected.
  • target: Name of the target.
  • CellComp: Subcellular location of target.

Data Access

The example data is hosted on Zenedo ().

Follow the commands below to download the example data.

1
2
3
4
work_dir=/path/to/work/directory
cd $work_dir
wget  https://zenodo.org/records/15701394/files/cosmxsmi_starter.raw.tar.gz
tar --strip-components=1 -zxvf cosmxsmi_starter.raw.tar.gz

Set Up the Environment

Define paths to all required binaries and resources, and target AWS S3 bucket. Optionally, specify a fixed color map for consistent rendering.

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
# ====
# Replace each placeholder with the actual path on your system.  
# ====
work_dir=/path/to/work/directory        # path to work directory that contains the downloaded input data
cd $work_dir

# Define paths to required binaries and resources
spatula=/path/to/spatula/binary         # path to spatula executable
punkst=/path/to/punkst/binary           # path to FICTURE2/punkst executable
tippecanoe=/path/to/tippecanoe/binary   # path to tippecanoe executable
pmtiles=/path/to/pmtiles/binary         # path to pmtiles executable
aws=/path/to/aws/cli/binary             # path to AWS CLI binary

# (Optional) Define path to color map. 
cmap=/path/to/color/map                 # Path to the fixed color map for rendering. cartloader provides a fixed color map at cartloader/assets/fixed_color_map_256.tsv.

# AWS S3 target location for cartostore
AWS_BUCKET="EXAMPLE_AWS_BUCKET"         # replace EXAMPLE_AWS_BUCKET with your actual S3 bucket name

# Activate the bioconda environment
conda activate BIOENV_NAME              # replace BIOENV_NAME with your bioconda environment name

Define data ID and analysis parameters:

1
2
3
4
5
6
7
8
# Unique identifier for your dataset
DATA_ID="cosmxsmi_hippo"                # change this to reflect your dataset name
PLATFORM="cosmx_smi"                    # platform information
SCALE=$(echo 1000/120|bc -l)              # scale from coordinate to micrometer

# LDA parameters
train_width=12                           # define LDA training hexagon width (comma-separated if multiple widths are applied)
n_factor=6,12                            # define number of factors in LDA training (comma-separated if multiple n-factor are applied)

How to Define Scaling Factors for CosMX SMI?

According to the README.html provided with the Pixel-seq dataset, each pixel has an edge length of 120 nm. To calculate the number of pixels per micrometer, use the formula: scale = 1000 / 120.

SGE Format Conversion

Convert the raw input to the unified SGE format. See more details in Reference page.

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
cartloader sge_convert \
  --makefn sge_convert.mk \
  --platform ${PLATFORM} \
  --in-csv ./input.tsv.gz \
  --units-per-um ${SCALE} \
  --out-dir ./sge \
  --exclude-feature-regex '^(BLANK|NegCon|NegPrb)' \
  --sge-visual \
  --spatula ${spatula} \
  --n-jobs 10
Parameter Required Type Description
--platform required string Platform (options: "10x_visium_hd", "seqscope", "10x_xenium", "bgi_stereoseq", "cosmx_smi", "vizgen_merscope", "pixel_seq", "generic")
--in-csv required string Path to the input TSV/CSV file
--units-per-um required float Scale to convert coordinates to microns (default: 1.0)
--out-dir required string Output directory for the converted SGE files
--makefn string File name for the generated Makefile (default: sge_convert.mk)
--exclude-feature-regex regex Pattern to exclude control features
--sge-visual flag Enable SGE visualization step (generates diagnostic image) (default: FALSE)
--spatula string Path to the spatula binary (default: spatula)
--n-jobs int Number of parallel jobs for processing (default: 1)

FICTURE analysis

Compute spatial factors using punkst (FICTURE2 mode). See more details in Reference page.

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
cartloader run_ficture2 \
  --makefn run_ficture2.mk \
  --main \
  --in-transcript ./sge/transcripts.unsorted.tsv.gz \
  --in-feature ./sge/feature.clean.tsv.gz \
  --in-minmax ./sge/coordinate_minmax.tsv \
  --cmap-file ${cmap} \
  --exclude-feature-regex '^(mt-.*$|Gm\d+$)' \
  --out-dir ./ficture2 \
  --width ${train_width} \
  --n-factor ${n_factor} \
  --spatula ${spatula} \
  --ficture2 ${punkst} \
  --n-jobs 10 \
  --threads 10
Parameter Required Type Description
--main required 1 flag Enable cartloader to run all five steps
--in-transcript required string Path to input transcript-level SGE file
--out-dir required string Path to output directory
--width required int or comma-separated list LDA training hexagon width(s)
--n-factor required int or comma-separated list Number of LDA factors
--makefn string File name for the generated Makefile (default: run_ficture2.mk )
--in-feature string Path to input feature file
--in-minmax string Path to input coordinate min/max file
--cmap-file string Path to color map file
--exclude-feature-regex regex Pattern to exclude features
--spatula string Path to the spatula binary (default: spatula)
--ficture2 string Path to the punkst directory (defaults to punkst repository within submodules directory of cartloader)
--n-jobs int Number of parallel jobs (default: 1)
--threads int Number of threads per job (default: 1)

1: cartloader requires the user to specify at least one action. Available actions includes: --tile to run tiling step; --segment to run segmentation step; --init-lda to run LDA training step; --decode to run decoding step; --summary to run summarization step; --main to run all above five actions.

cartloader Compilation

Generate pmtiles and web-compatible tile directories. See more details in Reference page.

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
cartloader run_cartload2 \
  --makefn run_cartload2.mk \
  --fic-dir ./ficture2 \
  --out-dir ./cartload2 \
  --id ${DATA_ID} \
  --spatula ${spatula} \
  --pmtiles ${pmtiles} \
  --tippecanoe ${tippecanoe} \
  --n-jobs 10 \
  --threads 10
Parameter Required Type Description
--fic-dir required string Path to the input directory containing FICTURE2 output
--out-dir required string Path to the output directory for PMTiles and web tiles
--id required string Dataset ID used for naming outputs and metadata
--makefn string File name for the generated Makefile (default: run_cartload2.mk)
--spatula string Path to the spatula binary (default: spatula)
--pmtiles string Path to the pmtiles binary (default: pmtiles)
--tippecanoe string Path to the tippecanoe binary (default: tippecanoe)
--n-jobs int Number of parallel jobs (default: 1)
--threads int Number of threads per job (default: 1)

Upload to Data Repository

AWS Uploads

Copy the generated cartloader outputs to your designated AWS S3 catalog path:

1
2
3
4
5
cartloader upload_aws \
  --in-dir ./cartload2 \
  --s3-dir "s3://${AWS_BUCKET}/${DATA_ID}" \
  --aws ${aws} \
  --n-jobs 10
Parameter Required Type Description
--in-dir required string Path to the input directory containing the cartloader compilation output
--s3-dir required string Path to the target S3 directory for uploading
--aws string Path to the AWS CLI binary
--n-jobs int Number of parallel jobs