SIGMA : Single-cell Identifier using Gaussian Mixture Approach

# SIGMA : Single-cell Identifier using Gaussian Mixture Approach SIGMA is a bioinformatics pipeline for automated identification of targeted cell types from multiple heterogeneous single-cell transcriptomic datasets (formats supported: `.h5ad`). It is designed to simplify and accelerate integrative large-scale single-cell analyses. ## Features - Quality control, filtering, and doublet removal - Extraction of target cell types based on user-defined gene markers - Harmonization and merging of multiple studies - Fully configurable via a single config file ## Installation Clone the repository: ```bash git clone git@github.com:Guitout31415/SIGMA.git cd SIGMA conda env create -f environment.yml conda activate SIGMA pip install snakemake ``` ## Configuration Edit `config_template.conf` to specify: - **[Metadata]**: Metadata fields to extract/conserve - **[Candidate]**: Candidate genes for initial filtering - **[Target]**: Positive marker genes for target cell identification - **[Exclude1]**: Negative or exclusion marker genes used to exclude cell types 1 - **[Exclude2]**: Negative or exclusion marker genes used to exclude cell types 2 - **[Thresholds]**: Filtering and assignment thresholds - **[Folder]**: Input/output folder paths - **[Options]**: Species, QC parameters, plotting options The number of components for the Gaussian Mixture Models (`n_components_target` and `n_components_exclu` in **[Thresholds]**) must be set explicitly to `kde` (KDE peak detection), `bic` (reproducible selection at the elbow of the BIC-vs-n_components curve), or a fixed integer. Both `kde` and `bic` are deterministic. With `bic`, a diagnostic BIC curve is saved to the plots folder. ## Usage Run the pipeline with Snakemake: ```bash snakemake \ --config file=ConfigFile.conf \ --cores 10 \ --jobs 5 \ --resources mem_mb=100000 ``` - Adjust `--cores` and `--jobs` as needed. - The main workflow is defined in `Snakefile`. ### Main Steps 1. **Quality control**: Filtering, outlier removal, doublet detection (`scripts/quality_control.py`) 2. **Target extraction**: Select cells expressing candidate/marker genes (`scripts/extract_target.py`) 3. **Harmonization**: Standardize metadata interface (`scripts/harmonize_metadata.py`) 4. **Merging**: Combine all studies into a single `.h5ad` (`scripts/merge_h5ad.py`) ## Output - Filtered and processed `.h5ad` files per study: `output_folder/qc/`, `output_folder/find/`, `output_folder/harmonized/` - Merged dataset: `output_folder/merge.h5ad` - Logs: `output_folder/logs/` - Plots: `output_folder/plots/` ## Troubleshooting - Check log files in `output_folder/logs/` for errors. - Ensure all dependencies are installed and input files are correctly formatted. ## License GPL-3.0 License. See [LICENSE](LICENSE).