SpaKnit: correlation subspace learning for integrating spatial multi-omics data

Integrating spatial multi-omics data presents significant challenges, particularly in uncovering the spatial patterns of cells and deciphering the real regulatory mechanisms among various omics. These insights are critical for harnessing the full potential of each modality while minimizing the impact of biotechnological biases that will lead to unstable results. Here, we introduce SpaKnit, a framework that treats multi-omics data as continuous functions mapped to spatial coordinates, enabling the discovery of nonlinear correlations among modalities. The effectiveness and robustness of SpaKnit are validated using simulated datasets and demonstrated across a range of tissue sections employing various techniques. Compared to existing methods, SpaKnit excels in identifying region-specific continuous spatial domains and maintains batch-consistency across trajectory inferences. By providing a novel perspective on the interplay between spatial information and multi-omics modalities, SpaKnit offers a flexible approach that can accommodate modality data of arbitrary dimensions.