orca

BioKERN: New AI Method Maps Tissue Images to Gene Expression by Preserving Biological Neighborhoods

Researchers Seungik Cho and Betul Orcan-Ekmekci introduced BioKERN, a biological kernel regularization approach for histology-to-transcriptomics neighborhood retrieval. Unlike existing methods that emphasize exact cross-modal matching, BioKERN learns representations that preserve biological neighborhood structure, allowing non-paired spots with shared molecular or spatial context to be correctly matched. This addresses a key limitation in spatially resolved biology, where current objectives often miss biologically meaningful correspondences. The method is detailed in a new arXiv paper and could improve the accuracy of spatial transcriptomics analyses.