Spatially Informed and Direction Sensitive Representation Learning for Spatial Transcriptomics

An arXiv preprint proposes Tessera, a spatial representation learning framework for spatial transcriptomics that addresses shortcomings in gene selection and modeling of cell-cell relationships. It combines spatially informed gene selection with a direction-preserving geometric inductive bias, using relative cell-cell relationships instead of absolute coordinates to achieve translation invariance while distinguishing direction. On held-out tissues and technologies, Tessera outperforms existing spatial foundation models in representation quality and generalizes zero-shot to contexts up to 16 times larger than those seen during training.