772-3d-OT: Geometry-Aware Framework for Spatial Alignment

Paper Talk · Apr 10, 2026 · 20m

The paper introduces 3d-OT, a sophisticated deep-learning framework designed to analyze and integrate spatial multi-omics data from biological tissues. By utilizing a geometry-aware architecture called PointNet++, the tool effectively identifies complex spatial domains and aligns disparate tissue slices, even when they suffer from nonrigid deformations. This computational approach outperforms existing methods in capturing fine-grained anatomical details, such as the specific layers of the mouse brain cortex. Furthermore, the framework enables the creation of 3D spatiotemporal trajectories…