What question did this study set out to answer?

The aim is to benchmark SpatialCNV tools for detecting copy-number variation in prostate cancer using both simulated and actual datasets.

January 22, 2026

Abstract A059: Benchmarking SpatialCNV in Prostate Cancer: tools, reference strategies, and workflows across simulated and real spatial transcriptomics

Key Points

The aim is to benchmark SpatialCNV tools for detecting copy-number variation in prostate cancer using both simulated and actual datasets.
Conducted a systematic benchmark across various spatial transcriptomics datasets.
Evaluated twelve CNV tools covering different methodologies for spatial analysis.
Assessed performance with and without histology for tumor detection and CNV calling.
CopyKAT achieved high accuracy in identifying tumor spots without histology using a reference-free mode.
inferCNV excelled in CNV calling using histologically benign epithelial references.
Developed an automated pipeline to define reference sets, demonstrating shared spatial clones in multiple patients.

Abstract

Abstract Background: Copy-number variation (CNV) is a hallmark of prostate cancer, including PTEN loss and TP53 alterations that shape tumor progression. Whole-genome spatial DNA sequencing technologies are not yet mature or widely accessible. Spatial transcriptomics-based CNV inference (SpatialCNV) produces genome-wide CNV maps and preserves histologic and spatial context. Unlike bulk DNA assays, it enables tracing of tumor subclones across tissue architecture. SpatialCNV can also identify tumor-containing spots when histology annotation is not provided, which is an essential feature for a lot of spatial cancer studies. Methods: We performed a systematic benchmark across simulated and real prostate spatial transcriptomics datasets, including sections with matched whole-exome truth and standard pathology. We evaluated twelve CNV tools, spanning single-cell-derived and spatial transcriptomics-specific methods. Performance was assessed in two scenarios: without histology (spot-level tumor detection) and with histology (CNV calling). We compared different strategies (stromal/immune, pure-benign epithelium, reference-free, etc) using correlation, sensitivity/specificity, and cross-patient robustness. Results: For detecting tumor-containing spots without histology, CopyKAT achieved the most reliable automatic identification when run in a reference-free mode after deconvolution-based selection of epithelial enriched spots. For CNV calling, inferCNV achieved the best performance when using a ‘pure-benign’ reference—histologically benign epithelial spots without CNV changes. We developed an automated pipeline to define pure-benign references and applied it to multi-patient cohorts, revealing spatially coherent clones shared between primary and metastatic tumors. Conclusions: SpatialCNV enables genome-wide characterization of prostate cancer clonality and metastatic relevant CNV events. Our benchmark provides pratical guidance and workflows, and supports understanding of prostate cancer clonality and spatial distribution. Citation Format: Jintong Shi. Benchmarking SpatialCNV in Prostate Cancer: tools, reference strategies, and workflows across simulated and real spatial transcriptomics abstract. In: Proceedings of the AACR Special Conference in Cancer Research: Innovations in Prostate Cancer Research and Treatment; 2026 Jan 20-22; Philadelphia PA. Philadelphia (PA): AACR; Cancer Res 2026;86 (2Suppl): Abstract nr A059.

Mark Helpful

Bookmark

Relay