SINTER3D: continuous 3D reconstruction of spatial transcriptomics via implicit neural representations
摘要
Spatial transcriptomics enables gene-expression profiling while preserving spatial context, but three-dimensional reconstruction from discrete tissue sections remains limited by large inter-section gaps and gene-wise independent interpolation. We develop SINTER3D, an implicit neural representation–based framework for joint three-dimensional interpolation of multiple genes. SINTER3D models gene expression as continuous functions of three-dimensional coordinates, enabling virtual section generation, spatial-domain identification, and cell-type deconvolution. Across datasets including adult mouse brain, human dorsolateral prefrontal cortex, developing human heart, Drosophila embryo, and breast cancer tissues, SINTER3D outperforms existing methods and reconstructs biologically meaningful three-dimensional molecular structures.