The abstract is read at the publisher; the summary is JClub's.
st2sc infers single-cell gene expression from multicellular spatial transcriptomics by combining Attention Gradient Flow Tracking with multiple-instance regression, improving instance segmentation and gene-count prediction on synthetic and Visium-like data.
- Why it matters: Spatial transcriptomics measures gene expression in spots that mix multiple cells, and recovering accurate per-gene expression for individual cells remains a key bottleneck.
- What they did: st2sc uses a Vision-Gene Transformer to generate attention maps for candidate cell instance detection via AGFT, then applies multiple-instance, multiple-target regressors to predict single-cell gene counts.
- The result: Evaluations on synthetic and Visium-like datasets show st2sc significantly outperforms prior methods for cellular instance segmentation and gene-count prediction, and it resolves contested 2um bins in Visium HD and recovers subtype marker-level analysis in human IPF lung.
The findingWhy it mattersWhat they didThe result
AI summary of the abstract · may contain errors · not medical advice · read the paper.