Integration and annotation of spatial multi-omic data with DIRAC.

DIRAC, a geometric deep learning model for spatial multi-omics, learns domain-invariant embeddings that enable shared cell-type annotation across sequencing and imaging datasets, including a new glioblastoma microvascular proliferation domain.

• Why it matters: Spatial multi-omic analysis is limited by two bottlenecks: integrating different molecular modalities and aligning cell types across studies with batch effects, modality differences, and site variation.
• What they did: DIRAC uses adaptive calibration in a generalized domain adaptation framework to learn task-discriminative features insensitive to domain shifts, then performs automated cell-type annotation via label transfer from reference datasets.
• The result: Across sequencing- and imaging-based datasets, DIRAC corrected cortical-layer annotations in severe Alzheimer’s disease, resolved mouse embryo structures, handled missing data and unseen cell types, and scaled to millions of cells.

AI summary of the abstract · may contain errors · not medical advice · read the paper.