Study Evaluates CNV Inference Methods Using scRNA-seq Data

Tumors are complex mosaics of genetically diverse cells, shaped by alterations such as copy number variations (CNVs), which involve the duplication or deletion of large DNA segments. These CNVs play a key role in cancer evolution, but detecting them at single-cell resolution is challenging. While single-cell RNA sequencing (scRNA-seq) offers a way to infer CNVs through gene expression data, the performance of available computational tools varies widely, creating uncertainty for researchers.

To address this, a multidisciplinary team led by Loma Linda University, the NCBI, and international partners conducted a systematic benchmarking study. Published in Precision Clinical Medicine, their work compared five popular CNV inference tools—HoneyBADGER, inferCNV, sciCNV, CaSpER, and CopyKAT—across a variety of scRNA-seq platforms and tumor models, including a new clinical dataset from a small cell lung cancer patient.

The researchers applied each method to datasets from different platforms, tumor-normal cell line pairs, artificial mixtures, and clinical samples. They assessed tool performance based on sensitivity, specificity, and accuracy in identifying tumor subpopulations. CaSpER and CopyKAT consistently provided the most balanced CNV inference, though their effectiveness depended on sequencing depth and platform. InferCNV and sciCNV were particularly strong in distinguishing tumor subclones within single-platform data.

The study also evaluated each tool’s ability to detect rare tumor populations. InferCNV showed high sensitivity when enough cells were sequenced, while sciCNV and HoneyBADGER were less effective. Batch effects posed challenges when combining data across platforms, but the allele-based version of HoneyBADGER was more robust to these issues. Clinical validation confirmed CaSpER and CopyKAT’s accuracy in CNV detection, and inferCNV and CopyKAT’s strength in identifying relapsed subclones.

"Understanding cancer at the single-cell level is essential for tackling tumor evolution and therapy resistance," said Charles Wang, co-corresponding author of the study. "Our benchmarking work provides the field with a clear reference point—highlighting not only which tools work best, but also under what conditions. It's a step toward more reliable and personalized cancer genomics."