SPICE: A Robust Computational Framework for Identifying Copy Number Variations in Spatial Transcriptomics

Abstract

Copy number variation (CNV), which alters the number of genomic segments, is a major driver of intratumor heterogeneity, characterized by spatially organized and genetically distinct cell populations. Recent advances in spatially resolved transcriptomic (SRT) technologies, which profile gene expression across thousands of spatially indexed tissue locations, offer a powerful opportunity to reconstruct the CNV architecture and dissect the spatial organization of cancer subclones. Here, we introduce SPICE (spatial inference of CNV events), a probabilistic method for identifying somatic CNVs and allele-specific copy number (ASCN) profiles from SRT data. A key feature of SPICE is its ability to integrate multiple complementary information available in SRT data, including gene expression, spatial coordinates, and heterozygous SNPs inferred from transcriptomic reads, to substantially enhance the accuracy and power of CNV detection. Using datasets generated across different SRT platforms, we first assess the reliability of SNPs derived from SRT data to ensure robust downstream inference. We then demonstrate that SPICE effectively integrates these modalities to deliver accurate and spatially coherent reconstruction of CNV landscapes and subclonal architecture, while maintaining excellent control of false discoveries. Together, SPICE provides a robust and effective solution for dissecting genomic heterogeneity in SRT studies of cancer.