Agentic AI enables cell segmentation-free cell-type inference and cell-type-guided cell segmentation in spatial transcriptomics

SSAF-T is a segmentation-free agentic framework that infers cell types directly from spatial gene expression and then enables accurate cell segmentation without predefined reference datasets.

• Why it matters: Cell-type identification in spatial transcriptomics often depends on cell segmentation, which can misassign transcripts at cell boundaries, while segmentation-free typing can be inaccurate and limit downstream analyses.
• What they did: SSAF-T performs reference-free cell typing using CeLLMan, which uses expert-level agents per organ to annotate cell types from a literature-grounded knowledge graph, and benchmarks on array-based and imaging-based datasets.
• The result: Across array-based and imaging-based spatial transcriptomics, SSAF-T is benchmarked for accuracy, reproducibility, and scalability, and its segmentation-free mapping provides accurate segmentation compatible with existing downstream pipelines for quantitative single-cell-level analysis.

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