Advances in spatially resolved technologies enable the characterization of tissues at molecular resolution by preserving spatial information. However, integrating and aligning spatial-omics data across different platforms and modalities remains challenging. Flexible tools for slice alignment, stitching and slice-to-volume 3D reconstruction are still lacking because available spatial-omics datasets are affected by partial overlapping, local non-rigid deformations, and large-scalability. Here we propose GEASO (Graph-based Elastic Alignment for Spatial-Omics data), a network-based algorithm for slice alignment, stitching and slice-to-volume 3D reconstruction. GEASO learns consistent spot features with graph neural network, and performs elastic registration to address rigid transformation and local deformation of slices by exploiting topological structure of spot connectivity graphs. GEASO also adopts acceleration strategies to enable its application to large-scale datasets. Experiment results demonstrate that GEASO outperforms state-of-the-art baselines in alignment, stitching and 3D reconstruction of slices across various platforms, modalities and tissues, providing a versatile tool for analyzing spatial-omics data.
Wang et al. (2026) studied this question.