windowCNV

windowCNV is a window-based tool for detecting copy number alterations (CNAs) from single-cell RNA-seq data. Inspired by infercnvpy, it extends its functionality with flexible CNA simulation, inference, and evaluation. The package supports cell type–aware analysis and event-level performance metrics.

Note: This package is experimental. Though benchmarked against inferCNVpy on simulated and real datasets, CNA classification accuracy in real-world datasets may be limited. Feedback and contributions are welcome.

---

Features

• Simulate CNAs globally or per cell type
• Infer CNAs from gene expression using customizable smoothing windows
• Assign CNA events to individual cells
• Evaluate precision, recall, and F1 score at the event level
• Visualize CNAs with heatmaps and summary tables

---

Installation

We recommend using a dedicated conda environment:

conda create -n windowcnv python=3.10
conda activate windowcnv

Then install windowCNV and its required dependencies:

pip install infercnvpy scanpy matplotlib pandas
pip install git+https://github.com/Li-Jiabei/windowCNV.git

---

Getting Started

Import the required packages:

import numpy as np
import pandas as pd
import scanpy as sc
import infercnvpy as cnv
import matplotlib.pyplot as plt
import warnings
from collections.abc import Sequence

import windowCNV as wcnv

---

Example Notebooks and Data

You can explore how to use windowCNV in the following notebooks:

Simulated CNAs (Benchmarking & Validation)

• Original infercnvpy usage: Shows the baseline workflow using infercnvpy, enhanced with our new plotting and evaluation functions.
• windowCNV implementation: Demonstrates the core windowCNV pipeline and comparison with infercnvpy.

These notebooks use the benchmarking dataset: PBMC_simulated_cnas_041025.h5ad

---

• CNA simulation and windowCNV application: Shows how to simulate CNAs and apply windowCNV inference.

This notebook uses the dataset: pbmc_10k_v3_filtered_feature_bc_matrix.h5

Note: Many real-world datasets (including the one above) lack chromosome and genomic position annotations in AnnData.var. To address this, we provide a helper function for automatic annotation. The usage is shown in the notebook. However, you must supply a gene annotation file.

In our example, we use the following reference file: mart_export_GRCh38.p14.txt This file contains:

• Gene stable ID
• Gene name
• Chromosome/scaffold name
• Gene start (bp)
• Gene end (bp)

The file was generated using Ensembl BioMart, which allows easy download of such annotations for genome build.

---

PSC scRNA-seq data with previously reported PSC CNAs

• windowCNV on PBMC-4k with CAS CNA Labels: Applies windowCNV to a 10x PBMC dataset with CAS-based high-confidence labels.

This notebook uses the dataset: SCP2745_high_conf_CAS_cell_types.h5ad

• windowCNV on TNBC iPSC-derived scRNA-seq data: Applies windowCNV to a triple-negative breast cancer (TNBC) iPSC dataset.

This notebook uses the dataset: GSM4476486_combined_UMIcount_CellTypes_TNBC1.txt.gz

• windowCNV on CRISPR-edited T cell iPSC data: Applies windowCNV to T cells derived from iPSCs edited with CRISPR.

This notebook uses the dataset: GSM7744300_GUIDEvsNT_CHR14_RESULTS.txt and gencode.v38.annotation.gtf