Abstract
Reconstructing the 3D spatial organization of transcription from 2D spatial data is a fundamental challenge in genomics. Here, we introduce Cytocraft, a computational framework that addresses this problem by inferring a shared, cell-type-specific 3D configuration of transcription centers from subcellular spatial transcriptomics profiles. After validating Cytocraft's robust accuracy in simulations (median relative error: 0.0346), we applied it to diverse datasets to explore spatial patterns that may reflect underlying transcriptional organization. In human nonsmall cell lung cancer, we observed consistent spatial repositioning patterns, suggesting that the marker MALAT1 may undergo directional translocation during malignant transformation across eight patient samples. In the developing axolotl brain, we found that transcription center reorganization exhibits conserved developmental dynamics across distinct cell types, suggesting coordinated developmental dynamics. By enabling 3D reconstruction of transcription center configurations from 2D data, Cytocraft provides a powerful tool to explore the spatial organization of transcription in development and disease.