Abstract A100: A computational framework for integrating multiscale spatial data in intraductal papillary mucinous neoplasms
Yang Liu, Akiko Sagara, Jimin Min, Tian Chu, Benson Chellakkan Selvanesan, Yunhe Liu, Aatur Singh, Anirban Maitra, Linghua WangAbstract
Background
Recent advances in spatial multi-omics have improved our ability to study tissue architecture, cellular heterogeneity, and cell-cell interactions within the tumor microenvironment. Xenium and Visium HD provide complementary strengths: Xenium offers true single-cell and subcellular spatial resolution but is limited by targeted gene panels, whereas Visium HD provides transcriptome-wide profiling at fixed spatial bins, making cell-level reconstruction challenging. Robust computational strategies are therefore needed to bridge these platforms and construct spatially resolved single-cell transcriptomic atlases of human tumors.
Methods
We developed a framework to integrate multiscale spatial transcriptomic data from Visium HD and Xenium in intraductal papillary mucinous neoplasms (IPMN). To convert Visium HD bin-level measurements into cell-level profiles, we benchmarked SMURF, Bin2cell, and the 10x Genomics bin-to-cell workflow. Methods were evaluated by reconstructed cell quality, transcript assignment, clustering structure, and downstream interpretability. SMURF was selected because it integrates nuclear segmentation with transcriptional similarity and generated the highest-quality cell-level profiles. We then tested cross-platform integration strategies, including Harmony, BBKNN, scANVI, and related approaches. Because conventional batch-correction methods showed limited ability to harmonize Visium HD and Xenium data due to differences in gene coverage, chemistry, and resolution, we adopted label transfer to leverage both platforms while preserving platform-specific spatial information.
Results
Our workflow reconstructed cell-level transcriptomic profiles from Visium HD data and enabled integration with Xenium datasets. Compared with alternative bin-to-cell methods, SMURF produced cells with improved transcriptomic quality, clearer clustering, and more reliable cell type annotation. Label transfer allowed Xenium-derived single-cell spatial annotations to be projected onto Visium HD-derived transcriptome-wide profiles, enabling spatially resolved characterization of tumor, stromal, and immune populations across IPMN tissues. This approach preserved the single-cell spatial precision of Xenium while expanding biological interpretation through whole-transcriptome coverage from Visium HD. The resulting atlas provided a more comprehensive view of cellular states and spatial organization within the IPMN microenvironment than either platform alone.
Conclusions
We present a computational framework for integrating Visium HD and Xenium data to construct a single-cell spatial transcriptomic atlas of IPMN. By benchmarking bin-to-cell reconstruction methods and evaluating cross-platform integration strategies, we show that SMURF-based cell reconstruction combined with label transfer bridges transcriptome-wide profiling and true single-cell spatial resolution. This framework provides a scalable strategy for studying tumor heterogeneity, microenvironmental organization, and cellular interactions in IPMN and other solid tumors.
Citation Format:
Yang Liu, Akiko Sagara, Jimin Min, Tian Chu, Benson Chellakkan Selvanesan, Yunhe Liu, Aatur Singh, Anirban Maitra, Linghua Wang. A computational framework for integrating multiscale spatial data in intraductal papillary mucinous neoplasms [abstract]. In: Proceedings of the AACR Conference on Pancreatic Cancer: New Frontiers in Biology and Therapeutic Development; 2026 Sep 25-28; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(18_Suppl_2):Abstract nr A100.