The paper introduces HoloTea, a novel computational framework designed to generate 3D volumetric tissue gene expression profiles from serial histology images. While traditional methods often analyze tissue slices in isolation, this model ensures anatomical continuity by retrieving morphologically similar data from adjacent sections to inform its predictions. The system utilizes a scalable flow-matching architecture and biology-aligned ZINB priors to accurately capture the complex nature of transcript count data. By leveraging global attention mechanisms, it remains efficient enough to process massive datasets that would otherwise overwhelm standard hardware. Experimental results across breast cancer and embryo datasets demonstrate that HoloTea significantly improves reconstruction accuracy over existing 2D and 3D baselines. Ultimately, this tool offers a cost-effective way to build virtual 3D tissues, facilitating a deeper understanding of cellular organization and disease.

References:

• Sanian M V, Hemmat A, Vahidi A, et al. 3D-Guided Scalable Flow Matching for Generating Volumetric Tissue Spatial Transcriptomics from Serial Histology[J]. arXiv preprint arXiv:2511.14613, 2025.