fastCNV: Fast and accurate copy number variation prediction from High-Definition Spatial Transcriptomics and scRNA-Seq Data

Background Predicting DNA copy number variations (CNVs) from spatial transcriptomics (ST), including Visium HD, or single-cell RNA-sequencing (scRNA-seq) data helps to distinguish malignant from non-malignant cells and to characterize the clonal architecture of tumor cells. Though there are existing methods of CNV analysis, they are often limited by slow speed, high memory consumption, lower accuracy in the absence of a reference for diploid cells, lower sensitivity at low read counts, and no support for clonal tree construction. Results To overcome these issues, we developed the R package fastCNV for detecting CNVs from ST, including Visium HD, or scRNA-seq data. FastCNV pools diploid references across samples and, within each sample, aggregates similar spots or cells with few reads into meta spots or cells. It automatically builds a clonality tree, and runs several times faster than other methods while using less memory. To measure the accuracy of fastCNV, we used 117 cancer cell line samples with both scRNA-seq and bulk whole-exome sequencing (WES) data. FastCNV identified CNVs highly correlated to those calculated from WES data (median correlation above 0.75), showing a significant improvement as compared to other methods such as inferCNV. Notably, fastCNV enables, for the first time, the analysis of CNVs from the Visium HD spatial transcriptomics technology. Applied to Visium HD breast cancer ST data, fastCNV identifies tumor subclones tightly related to different histologies, linking specific genetic aberrations to tumor progression. Conclusions FastCNV is a significant improvement on existing R methods for CNV detection from ST, including Visium HD, or scRNA-seq data in terms of speed, memory usage, sensitivity and accuracy. This highlights its potential to advance cancer research and personalized medicine. FastCNV is available at https://github.com/must-bioinfo/fastCNV/ .