Abstract
Spatial transcriptomics alignment is challenged by technical variations, including geometric distortions from tissue preparation and platform-driven differences in resolution and modality. These issues create diverse alignment scenarios, from matched and mismatched resolutions to cross-modality integration, while partial tissue coverage further complicates the task. To overcome these limitations, we introduce GALA (Genetic Algorithm-guided Large Deformation Alignment), a unified, landmark-free framework that couples global affine transformation and local diffeomorphic deformation within a single optimization. Its modality-aware rasterization harmonizes transcriptomic and histological data into a shared grid, enabling landmark-free, multimodal alignment across resolutions, and modalities. Evaluated on diverse human and mouse datasets, GALA outperforms existing methods in accuracy, computational efficiency, and biological interpretability for both complete and partial tissue alignment.