Bayesian segmentation of imaging-based spatial transcriptomics data

• News ([0.6.0] — 2023-04-13)
• Overview
• Installation
  • Binary download
  • Install as a Julia package
  • Docker
• Run
  • Dataset preview
  • Full run
    • Normal run
    • Using a prior segmentation
      • Segmenting stains
    • Segmenting cells with pronounced intracellular structure
    • Outputs
    • Choice of parameters
  • Extract NCVs (neighborhood composition vectors)
  • Multi-threading
• Advanced configuration
• Preparing a release
• Citation

A major update!

• Simplified installation
• Fixed various issues: memory usage of NCVs, polygon output
• Finished loom support
• Updated CLI and configuration (breaking changes in the config file structure!)
• And more

See the changelog for more detalis.

Baysor is a tool for performing cell segmentation on imaging-based spatial transcriptomics data. It optimizes segmentation considering the likelihood of transcriptional composition, size and shape of the cell. The approach can take into account nuclear or cytoplasm staining, however, can also perform segmentation based on the detected molecules alone. The details of the method are described in the paper, or pre-print (old version of the text). To reproduce the analysis from the paper see BaysorAnalysis repo.

See the 16-min live-demo of Baysor for an overview of the workflow! Also, here is my 2023 video with the paper presentation and some updates on the ideas.

Do you have any question? Start a discussion!

The easiest way to install Baysor on Linux is to download a binary from the release section (see Assets). There, you can use bin/baysor executable. For other platforms, "Install as a Julia package" is a recommended way. If you know how to reliably compile binaries for MacOS or Windows, please, let me know in issues or over email!

If you need to install julia, juliaup is a recommended way. It's cross-platform and doesn't require admin privileges. TL;DR: curl -fsSL https://install.julialang.org | sh .

To install Baysor as a Julia package run the following command from your CLI (it requires gcc or clang installed):

julia -e 'using Pkg; Pkg.add(PackageSpec(url="https://github.com/kharchenkolab/Baysor.git")); Pkg.build()'

It will install all dependencies, compile the package and create an executable in ~/.julia/bin/baysor. This executable can be moved to any other place if you need it.

Alternatively, you can use Docker. It contains executable baysor to run Baysor from CLI, as well as IJulia installation to use Baysor with Jupyter. The repo also has images for older versions.

docker run -it --rm vpetukhov/baysor:latest

Build by hands:

docker pull julia:latest
cd Baysor/docker
docker build .

You can find more info about dockers at Docker Cheat Sheet.

As a full run takes some time, it can be useful to run a quick preview to get meaning from the data and to get some guesses about the parameters of the full run.

baysor preview [-x X_COL -y Y_COL --gene GENE_COL -c config.toml -o OUTPUT_PATH] MOLECULES_CSV

Here:

• MOLECULES_CSV must contain information about x and y positions and gene assignment for each molecule
• Parameters X_COL, Y_COL and GENE_COL must specify the corresponding column names. Default values are "x", "y" and "gene" correspondingly
• OUTPUT_PATH determines the path to the output html file with the diagnostic plots

To run the algorithm on your data, use

baysor run [-s SCALE -x X_COL -y Y_COL -z Z_COL --gene GENE_COL] -c config.toml MOLECULES_CSV [PRIOR_SEGMENTATION]

Here:

• MOLECULES_CSV must contain information about x and y positions and gene assignment for each molecule
• PRIOR_SEGMENTATION is optional molecule segmentation obtained from another method (see Using prior segmentation)
• Parameters X_COL, Y_COL, Z_COL and GENE_COL must specify the corresponding column names. Default values are "x", "y", "z" and "gene" correspondingly.
• SCALE is a crucial parameter and must be approximately equal to the expected cell radius in the same units as "x", "y" and "z". In general, it's around 5-10 micrometers, and the preview run can be helpful to determine it for a specific dataset (by eye, for now).

To see the full list of command-line options run

baysor run --help

For more info see examples (though probably are out-of-date).

For the description of all config parameters, see example_config.toml.

In some cases, you may want to use another segmentation as a prior for Baysor. The most popular case is having a segmentation based on DAPI/poly-A stainings: such information helps to understand where nuclei are positioned, but it's often quite imprecise. To take this segmentation into account you can pass it as the second positional argument to Baysor:

baysor run [ARGS] MOLECULES_CSV [PRIOR_SEGMENTATION]

Here, PRIOR_SEGMENTATION can be a path to a binary image with a segmentation mask, an image with integer cell segmentation labels or a column name in the MOLECULES_CSV with integer cell assignment per molecule (0 value means no assignment). In the latter case, the column name must have : prefix, e.g. for column cell you should use baysor run [ARGS] molecules.csv :cell. In case the image is too big to be stored in the tiff format, Baysor supports MATLAB '.mat' format: it should contain a single field with an integer matrix for either a binary mask or segmentation labels. When loading the segmentation, Baysor filters segments that have less than min-molecules-per-segment molecules. It can be set in the toml config, and the default value is min-molecules-per-segment = min-molecules-per-cell / 4. Note: only CSV column prior is currently supported for 3D segmentation.

To specify the expected quality of the prior segmentation you may use prior-segmentation-confidence parameter. The value 0.0 makes the algorithm ignore the prior, while the value 1.0 restricts the algorithm from contradicting the prior. Prior segmentation is mainly needed for the cases where gene expression signal is not enough, e.g. with very sparse protocols (such as ISS or DARTFISH). Another potential use case is high-quality data with a visible sub-cellular structure. In these situations, setting prior-segmentation-confidence > 0.7 is recommended. Otherwise, the default value 0.2 should work well.

If you have a non-segmented DAPI image, the simplest way to segment it would go through the following steps ImageJ:

1. Open the image (File -> Open)
2. Go to Image -> Type and pick "8-bit"
3. Run Process -> Filters -> Gaussian Blur, using Sigma = 1.0. The value can vary, depending on your DAPI, but 1.0 generally works fine.
4. Run Image -> Adjust -> Auto Threshold, using Method = Default. Different methods can give the best results for different cases. Often "Mean" also works well.
5. Run Process -> Binary -> Watershed
6. Save the resulting image in the .tif

Another promising tool is CellPose, however, it may require some manual labeling to fine-tune the network.

High-resolution protocols, such as MERFISH or seq-FISH, can capture the intracellular structure. Most often, it would mean a pronounced difference between nuclear and cytoplasmic gene composition. By default, such differences would push Baysor to recognize compartments as different cells. However, if some compartment-specific genes are known, they may be used to mitigate the situation. These genes can be specified through --config.segmentation.nuclei-genes and --config.segmentation.cyto-genes options, e.g.:

baysor run -m 30 --n-clusters=1 -s 30 --scale-std=50% --config.segmentation.nuclei-genes=Neat1 --config.segmentation.cyto-genes=Apob,Net1,Slc5a1,Mptx2 --config.data.exclude-genes='Blank*' ./molecules.csv

Please, notice that it's highly recommended to set --n-clusters=1, so molecule clustering would not be affected by compartment differences.

Note. Currently, there is no automated way to determine such compartment-specific genes. So, the only way we can suggest is interactive explaration of data. In theory, it should be straightforward to infer such information from DAPI and poly-A stains, however, it is not implemented yet. If you have a particular need for such functionality, please submit an issue with the description of your experimental setup.

Segmentation results:

• segmentation_counts.loom or segmentation_counts.tsv (depends on --count-matrix-format): count matrix with segmented stats. In the case of loom format, column attributes also contain the same info as segmentation_cell_stats.csv.
• segmentation.csv: segmentation info per molecule:
  • confidence: probability of a molecule to be real (i.e. not noise)
  • cell: id of the assigned cell. Value "" corresponds to noise.
  • cluster: id of the molecule cluster
  • assignment_confidence: confidence that the molecule is assigned to a correct cell
  • is_noise: shows whether molecule was assigned to noise (it equals true if and only if cell == "")
  • ncv_color: RGB code of the neighborhood composition coloring
• segmentation_cell_stats.csv: diagnostic info about cells. The following parameters can be used to filter low-quality cells:
  • area: area of the convex hull around the cell molecules
  • avg_confidence: average confidence of the cell molecules
  • density: the number of molecules in a cell divided by the cell area
  • elongation: ratio of the two eigenvalues of the cell covariance matrix
  • n_transcripts: number of molecules per cell
  • avg_assignment_confidence: average assignment confidence per cell. Cells with low avg_assignment_confidence have a much higher chance of being an artifact.
  • max_cluster_frac (only if n-clusters > 1): fraction of the molecules coming from the most popular cluster. Cells with low max_cluster_frac are often doublets.
  • lifespan: number of iterations the given component exists. The maximal lifespan is clipped proportionally to the total number of iterations. Components with a short lifespan likely correspond to noise.

Visualization:

• segmentation_polygons.json: polygons used for visualization in GeoJSON format. In the case of 3D segmentation, it is an array of GeoJSON polygons per z-plane, as well as "joint" polygons. Shown only if --save-polygons=geojson is set.
• segmentation_diagnostics.html: visualization of the algorithm QC. Shown only when -p is set.
• segmentation_borders.html: visualization of cell borders for the dataset colored by local gene expression composition (first part) and molecule clusters (second part). Shown only when -p is set.

Other:

• segmentation_params.dump.toml: aggregated parameters from the config and CLI

Most important parameters:

• scale is the most sensitive parameter, which specifies the expected radius of a cell. It doesn't have to be precise, but the wrong setup can lead to over- or under-segmentation. This parameter is inferred automatically if cell centers are provided.
• min-molecules-per-cell is the number of molecules, required for a cell to be considered as real. It really depends on the protocol. For instance, for ISS it's fine to set it to 3, while for MERFISH it can require hundreds of molecules.

Some other sensitive parameters (normally, shouldn't be changed):

• new-component-weight is proportional to the probability of generating a new cell for a molecule, instead of assigning it to one of the existing cells. More precisely, the probability to assign a molecule to a particular cell linearly depends on the number of molecules, already assigned to this cell. And this parameter is used as the number of molecules for a cell, which is just generated for this new molecule. The algorithm is robust to small changes in this parameter. And normally values in the range of 0.1-0.9 should work fine. Smaller values would lead to slower convergence of the algorithm, while larger values force the emergence of a large number of small cells on each iteration, which can produce noise in the result. In general, the default value should work well.

Run parameters:

• --config.segmentation.n-cells-init expected number of cells in data. This parameter influence only the convergence speed of the algorithm. It's better to set larger values than smaller ones.
• --config.segmentation.iters number of iterations for the algorithm. At the moment, no convergence criteria are implemented, so it will work exactly iters iterations. Thus, too small values would lead to non-convergence of the algorithm, while larger ones would just increase working time. Optimal values can be estimated by the convergence plots, produced among the results.

To extract NCVs you need to run:

baysor segfree [-x X_COL -y Y_COL --gene GENE_COL -c config.toml -o OUTPUT_PATH] MOLECULES_CSV

The results will be stored in loom format with /matrix corresponding to the NCV matrix and /col_attrs/ncv_color showing the NCV color.

All running options support some basic multi-threading. To enable it, set JULIA_NUM_THREADS environment variable before running the script You can either do it globally by running export JULIA_NUM_THREADS=13 or for an individual command:

JULIA_NUM_THREADS=13 baysor run -m 30 -s 30 ./molecules.csv

The pipeline options are described in the CLI help. Run baysor --help or the corresponding command like baysor run --help for the list of main options.

However, there are additional parameters that can be specified through the TOML config. See example_config.toml for their description. All parameters from the config can also be passed through the command line. For example, to set exclude_genes from the data section you need to pass --config.data.exclude-genes='Blank*,MALAT1' parameter. Please, keep in mind that the CLI parameters require replacing all underscores (_) with -.

For more details on the syntax for CLI arguments, see the Comonicon documentation. TL;DR, possible spelling options are: --x-column X or -x X, also --x-column=X or -nX. And if you're using strings with unusual symbols like * or ?, it's better to have them in quotes: --config.data.exclude-genes='Blank*'.

Update the version in the Project.toml, then:

export BAYSOR_VERSION=v0.6.0
# ...change the version in the Project.toml...
LazyModules_lazyload=false julia --project ./deps/build.jl app
# ...test transferability...
zip -r "baysor-x86_x64-linux-${BAYSOR_VERSION}_build.zip" LICENSE README.md ./bin/baysor/*

git push origin master
docker build -t vpetukhov/baysor:latest -t "vpetukhov/baysor:$BAYSOR_VERSION" --build-arg CACHEBUST=$(date +%s) .
git tag -a $BAYSOR_VERSION -m $BAYSOR_VERSION
git push origin master --tags

docker push -a vpetukhov/baysor

If you find Baysor useful for your publication, please cite:

Petukhov V, Xu RJ, Soldatov RA, Cadinu P, Khodosevich K, Moffitt JR & Kharchenko PV.
Cell segmentation in imaging-based spatial transcriptomics.
Nat Biotechnol (2021). https://doi.org/10.1038/s41587-021-01044-w