Three-dimensional spatial transcriptomics at isotropic resolution enabled by artificial intelligence

isoST is a generative model designed to reconstruct 3D spatial transcriptomic profiles with isotropic resolutions from sparsely sampled serial sections. Accurately mapping isotropic-resolution 3D spatial transcriptomes is a major challenge in biology. Current technologies cannot directly achieve full 3D profiling, so tissues are typically sectioned into serial 2D slices for individual profiling. We present <b>isoST</b>, a framework to reconstruct continuous, isotropic-resolution 3D transcriptomic landscapes from sparsely sampled serial sections. Assuming gene expression varies smoothly in 3D space, isoST models expression dynamics along tissue depth using stochastic differential equations (SDEs), producing a continuous 3D field that enables high-fidelity reconstruction from limited slices.isoST（isoST）是一款生成式模型，旨在从稀疏采样的连续组织切片中重建具备各向同性分辨率的三维空间转录组图谱。精准构建各向同性分辨率的三维空间转录组图谱是生物学领域的重大挑战。当前技术无法直接完成全三维空间转录组谱分析，因此通常需将组织切割为连续二维切片以分别开展转录组谱分析。本研究提出**isoST**，一款可从稀疏采样的连续组织切片中重建连续、各向同性分辨率的三维转录组图谱的分析框架。该模型假设基因表达在三维空间中呈平滑变化，借助随机微分方程（stochastic differential equations，SDEs）对沿组织深度的基因表达动态进行建模，生成连续三维场，从而可从有限切片中实现高保真的三维转录组图谱重建。