scMalignantFinder: Distinguishing Malignant Cells in Single-Cell and Spatial Transcriptomics by Leveraging Cancer Signatures
scMalignantFinder is a Python package designed for analyzing cancer single-cell RNA-seq and spatial transcriptomics datasets to distinguish malignant cells from their normal counterparts. Trained on over 400,000 high-quality single-cell transcriptomes, scMalignantFinder uses curated pan-cancer gene signatures for training set calibration and selects features by taking the union of differentially expressed genes across each dataset.
- Enhanced downstream analysis capabilities with a new module for cancer cell state analysis
Expanded input file format compatibility
- Extended support for test data input to include
.txt,.tsv, and.csvfiles, as well as their compressed.gzversions.
New spatial region identification feature
- Add the ability to identify malignant regions from spatial transcriptomics data by applying a clustering-based method.
Enhanced Flexibility for Test Input
- Test data can now be provided as a path to an .h5ad file or directly as an AnnData object.
New features
- Introduced malignancy probability output
We recommend using a conda environment to install scMalignantFinder.
- Create and activate a conda environment
conda create -n scmalignant python=3.10.10
conda activate scmalignant- Install
scMalignantFinderfrom PyPI:
pip install scMalignantFinderOptional: scMalignantFinder includes a built-in pan-cancer cell type annotation tool, scATOMIC. If you want to perform basic cell type annotation before identifying malignant cells, follow the scATOMIC official tutorial to complete its installation in the same conda environment.
A pretrained model and a list of ordered features are provided in the model directory:
- Pretrained model:
model/model.joblib - Ordered feature list:
model/ordered_feature.tsv
Users can also download or use the training data for training the model.
- Training data: Download the training data used in the original study from here, or use your own dataset to train the model.
- Feature file: The feature list file can be collected from here.
- Example test data:
In addition, for malignant region identification in spatial transcriptomics data, the malignant cell gene signature file is provided as model/sc_malignant_deg.gmt.
Alternatively, users can supply their own gene signature file, provided it follows the same .gmt format.
### Load package
from scMalignantFinder import classifier
# Initialize model
model = classifier.scMalignantFinder(
test_input="path/to/test_data.h5ad", # Test data input. Supports:
# - AnnData object
# - Path to .h5ad file
# - Path to tab-delimited .txt/.tsv file
# - Path to comma-delimited .csv file
# (.txt, .tsv, and .csv can also be gzip-compressed as
# .txt.gz, .tsv.gz, or .csv.gz)
# For text-based files (.txt, .tsv, .csv),
# rows must correspond to gene symbols
# and columns to cell barcodes.
pretrain_dir=None, # Directory containing pretrained model and feature list.
# If None, the model will be trained from scratch. Default: None.
train_h5ad_path="/path/to/train_data.h5ad", # Path to training dataset (.h5ad)
feature_path="/path/to/features.txt", # Path to ordered feature list (.txt)
model_method="LogisticRegression", # Classifier method: LogisticRegression, RandomForest, or XGBoost
norm_type=True, # Whether to normalize test data. Default: True.
n_thread=1 # Number of threads for parallel processing
)
# Load data
model.load()
# Predict malignancy
result_adata = model.predict()
# View results
print(result_adata.obs["scMalignantFinder_prediction"].head())
## Example output for scMalignantFinder_prediction:
Index
KUL01-T_AAACCTGGTCTTTCAT Malignant
KUL01-T_AAACGGGTCGGTTAAC Malignant
KUL01-T_AAAGATGGTATAGGGC Normal
KUL01-T_AAAGATGGTGGCCCTA Malignant
KUL01-T_AAAGCAAGTAAACACA Malignant
Name: scMalignantFinder_prediction, dtype: category
Categories (2, object): ['Normal', 'Malignant']
print(result_adata.obs["malignancy_probability"].head())
## Example output for malignancy_probability:
Index
KUL01-T_AAACCTGGTCTTTCAT 0.98578
KUL01-T_AAACGGGTCGGTTAAC 0.78968
KUL01-T_AAAGATGGTATAGGGC 0.243564
KUL01-T_AAAGATGGTGGCCCTA 0.8796
KUL01-T_AAAGCAAGTAAACACA 0.6598
Name: malignancy_probability, dtype: float64On top of the malignancy probability from Identify malignant cells from scRNA-seq data, malignant regions in spatial transcriptomics data can be further identified by integrating gene signatures and image-based features:
from scMalignantFinder import spatial, utils # assuming spatial module provides relevant methods
# Step 1: Calculate AUCell score using scRNA-seq-derived gene sets
sc_gmt = './model/sc_malignant_deg.gmt'
adata = utils.aucell_cal(adata, sc_gmt)
# Step 2: Extract image-based features from spatial images
adata = spatial.image_cal(adata)
# Step 3: Integrate multi-modal features to segment malignant regions
adata = spatial.region_identification(
adata,
features=['malignancy_probability', 'Malignant_up', 'image_score'],
# A list of feature names from adata.obs used for clustering.
# These features are used to compute pairwise distances and perform hierarchical clustering.
nclus=3,
# Number of clusters to define from hierarchical clustering (default: 3).
# One of the clusters will be identified as "Malignant", and the others as "Normal".
define_feature='Malignant_up',
# The key feature used to determine which cluster corresponds to malignant regions.
# The cluster with the highest average value of this feature will be labeled as "Malignant".
spatial_nn=True
# Whether to refine region labels using spatial neighbor information (default: True).
# If True, each spot's label may be adjusted by majority vote among its spatial neighbors.
)
# Example output for scMalignantFinder_prediction:
print(adata.obs[['cluster','region_prediction']])
cluster region_prediction
AAACAAGTATCTCCCA-1 0 Normal
AAACACCAATAACTGC-1 1 Malignant
AAACAGAGCGACTCCT-1 2 Normal
AAACAGGGTCTATATT-1 0 Normal
AAACAGTGTTCCTGGG-1 1 MalignantTo support downstream functional interpretation of malignant cells, we collected and curated 67 cancer cell state gene sets from a pan-cancer study. These gene sets represent a wide spectrum of cancer-associated cellular programs (e.g., cell cycle, EMT, immune evasion, hypoxia) across multiple cancer types.
You can quantify the enrichment of these gene sets in individual cells using AUCell scoring:
from scMalignantFinder import utils
# Path to the pan-cancer curated gene sets
pan_cancer_gene_sets = "/path/to/model/Malignant_MPs.Gavish_2023.gmt"
# Compute AUCell scores for each gene set
adata = utils.aucell_cal(adata, pan_cancer_gene_sets, norm_type=False)
# View results
print(adata.obs.loc[:,adata.obs.columns.str.startswith('MP')].iloc[:5,:3])
MP1 Cell Cycle - G2/M MP2 Cell Cycle - G1/S MP3 Cell Cylce HMG-rich
KUL01-T_AAACCT 0.045819 0.000000 0.306887
KUL01-T_AAACGG 0.155027 0.078003 0.227548
KUL01-T_AAAGAT 0.000000 0.000000 0.293480
KUL01-T_AAAGAG 0.000000 0.000000 0.239118
KUL01-T_AAAGCA 0.068728 0.000000 0.272176
If you use scMalignantFinder in your research, please cite:
Qiaoni, Yu, et al. "scMalignantFinder distinguishes malignant cells in single-cell and spatial transcriptomics by leveraging cancer signatures.", Communications Biology, 2025. DOI: https://doi.org/10.1038/s42003-025-07942-y
BibTeX:
@article{yu2025scmalignantfinder,
title={scMalignantFinder distinguishes malignant cells in single-cell and spatial transcriptomics by leveraging cancer signatures},
author={Yu, Qiaoni and Li, Yuan-Yuan and Chen, Yunqin},
journal={Communications Biology},
volume={8},
number={1},
pages={504},
year={2025},
publisher={Nature Publishing Group UK London}
}