The paper introduces STAIR, a novel computational framework designed to reconstruct three-dimensional (3D) biological atlases from two-dimensional spatial transcriptomics (ST) data. Traditional methods often struggle with batch effects and unknown physical distances between tissue slices, but STAIR addresses these issues using a heterogeneous graph attention network that adaptively integrates molecular and spatial features. This end-to-end tool is uniquely capable of unsupervised z-axis reconstruction, allowing researchers to determine the relative positioning of slices without prior anatomical measurements. Beyond 3D construction, STAIR facilitates 2D alignment and the integration of new tissue sections into existing reference frameworks across different samples and sequencing platforms. Benchmarking across diverse datasets, including the mouse brain and human breast cancer tissues, demonstrates its superior accuracy in identifying spatial domains and capturing complex biological heterogeneity. Ultimately, STAIR provides a robust solution for visualizing organ-level architecture and understanding the spatial molecular drivers of tissue organization.

References:

• Yu Y, Xie Z. Spatial transcriptomic alignment, integration, and 3D reconstruction by STAIR[J]. Genome Biology, 2025, 26(1): 427.