Cellist

Cell identification in high-resolution Spatial Transcriptomics

Cellist is a computational method to perform cell segmentation on high-resolution spatial transcriptomics (ST) data, including sequncing-based (e.g. Stereo-seq and Seq-Scope) and imaging-based (e.g. seqFISH+ and STARmap) technologies.

Change Log

v0.0.1a

• Build Cellist.

v1.0.0

• Release Cellist.

v1.1.0

• Update Cellist model.
• Support Cellpose for image-based segmentation.

v1.1.1

• Fix a typo in the Watershed segmentation.
• Update package dependencies.

v1.2.0

• cellist seg: new --count-layers option (default auto). Extra count columns of the GEM such as EXONIC / INTRONIC are aggregated into additional cell-level matrices ({outprefix}_Cellist_segmentation_cell_count_EXONIC.h5, ..._INTRONIC.h5) with the same gene and cell ordering as the main count matrix, so they can be used directly as the spliced / unspliced layers for RNA velocity.
• cellist seg: default model parameters changed to --cell-radius 25, --alpha 0.9, --beta 5.0 (v1.1.1: 15, 0.8, 10). Runs that relied on the old defaults should pass them explicitly.
• cellist seg: the merge of overlapping slices / patches is now deterministic. Each spot keeps the assignment with the highest probability over all slices; previously a first-occurrence de-duplication ran before the probability sort, so about 1% of labels depended on the order in which worker results were collected and changed between otherwise identical runs.
• cellist seg: memory-bounded merge that streams the GEM with polars; the GEM count column name (MIDCount / MIDCounts) is detected automatically.
• cellist align: new --transpose, --transpose_auto and --prealigned options. cellist watershed: new --offset and --test-mode options. cellist cellpose: GPU device selected automatically.
• New dependency polars.

System Requirements

Hardware Requirements

For optimal performance, especially when using Cellpose for initial image-based segmentation, a system with a GPU is recommended.

Operating System

This package is supported on Linux systems and has been tested on Rocky Linux release 9.4.

Python Rependencies

The core dependencies include:

scikit-learn>=1.3.1
scikit-image>=0.20.0
h5py>=3.8.0
tables>=3.7.0

A complete list of required packages is available in requirements.txt.

Install Cellist

Step-by-step Instructions

We recommend installing Cellist in a fresh conda environment:

git clone https://github.com/wanglabtongji/Cellist.git
cd Cellist
conda create -n Cellist python=3.10
conda activate Cellist
pip install -r requirements.txt
pip install .

Note: After installation, users can verify the installation by running cellist --help. The command may return a message such as fastpd is not installed. Please compile the fastpd library. This warning originates from Spateo and does not affect the functionality of Cellist.

Typical Install Time

On a high-performance server, installation typically takes 10 minutes.

Documentation

For full installation and usage of Cellist, please refer to the documentation.

Usage

cellist --help
usage: cellist [-h] [-v] {seg,align,watershed,cellpose,impute} ...

Cellist (Cell identification in high-resolution Spatial Transcriptomics) is a cell segmentation tool for high-
resolution spatial transcriptomics.

positional arguments:
  {seg,align,watershed,cellpose,impute}
    seg                 Run Cellist segmentation on high-resolution spatial transcriptomics.
    align               Refine alignment between image and spatial transcriptomics.
    watershed           Run initial watershed segmentation on the staining image.
    cellpose            Run initial cellpose segmentation on the staining image.
    impute              Perform spatially-aware gene imputation within each cluster.

optional arguments:
  -h, --help            show this help message and exit
  -v, --version         Print version info.