A single-cell modeling framework that redefines SNVs as endogenous perturbations of cellular state.
Single-nucleotide variants (SNVs) are central to tumor evolution, yet their functional consequences remain largely unresolved at single-cell resolution. Crucially, it remains unknown how the impact of specific SNVs varies across patients, cell types, or cellular states — hindering mechanistic understanding and therapeutic stratification. Current strategies operate predominantly at the bulk level and depend on population recurrence or evolutionary constraint, capturing signatures of long-term selection rather than direct, acute cellular effects.
PrismSNV redefines SNVs as endogenous perturbations of cellular state. By quantifying mutation-induced displacement in transcriptomic space, PrismSNV directly measures functional impact and uncovers the dynamic roles of individual SNVs throughout the spatiotemporal evolution of tumors.
| Feature | Description |
|---|---|
| 🎯 Single-cell resolution | Measures SNV functional impact directly at the single-cell level, rather than relying on bulk aggregates. |
| 🧪 Endogenous perturbation modeling | Treats each SNV as a natural perturbation of cellular state, capturing acute effects instead of long-term selection signals. |
| 📐 Transcriptomic displacement | Quantifies mutation-induced displacement in transcriptomic space as a direct readout of functional impact. |
| 🕰️ Spatiotemporal dynamics | Uncovers how individual SNVs change roles throughout tumor evolution. |
This guide installs PrismSNV from a local checkout without publishing it to conda.
Clone the repository and enter the project directory:
git clone https://github.com/xjtu-omics/PrismSNV.git
cd PrismSNVconda create -n prismsnv python=3.10 -y
conda activate prismsnvconda install -c conda-forge -c bioconda bash samtools bedtools openjdk -y
⚠️ You also need a VarScan JAR file and should pass it with--varscan-jar.
Run this from the repository root:
pip install -e .💡 Use
pip install .instead if you want a non-editable install.
prismsnv --helpPrismSNV is driven by a single prismsnv command-line entry point that dispatches to five subcommands.
| Command | Description |
|---|---|
bam2vcf |
Runs the Bash SNV-calling pipeline for one or more BAM files, producing filtered VCF files after removing RNA-editing sites. |
snv2barcode |
Builds per-sample and merged barcode-by-SNV AnnData matrices from BAM, VCF, and barcode inputs defined in a YAML config. |
pre_train |
Aligns pretraining and finetuning RNA AnnData inputs, then trains the RNA-only backbone model. |
snv_effect |
Trains or evaluates the SNV perturbation model and exports functional-effect results. |
get_template |
Writes a train_config.yaml template into the current directory. |
# 0. Generate a configuration template
prismsnv get_template --output train_config.yaml
# 1. Call SNVs from BAM files
prismsnv bam2vcf --outer-jobs 6 --inner-threads 4 \
--reference genome.fa --varscan-jar VarScan.jar \
--rna-edit-bed RNA_edit.bed --out-dir ./snv_call_out \
--bam-files sample1.bam sample2.bam
# 2. Build barcode-by-SNV matrices
prismsnv snv2barcode snv2barcode_config.yaml
# 3. Train the RNA backbone model
prismsnv pre_train -y train_config.yaml
# 4. Train/evaluate the SNV perturbation model
prismsnv snv_effect --n_gpu 3 -y train_config.yamlInspect any subcommand without processing data:
prismsnv <command> --helpPlease see the PrismSNV documentation for detailed usage.
If you encounter any issues during use, please try updating PrismSNV to the latest version. If the issue persists, feel free to submit it on the issue page or contact us directly:
| Author | X | |
|---|---|---|
| Peisen Sun | sunpeisen@stu.xjtu.edu.cn | @Sun_python |
| Kai Ye | kaiye@xjtu.edu.cn | — |
This project is licensed under the GNU General Public License v3.0 (GPLv3).
