Overview
Imaging-based spatial transcriptomics (e.g., Xenium, MERFISH, CosMx) localizes individual transcripts in tissue, but assigning those transcripts to cells depends on segmentation that can produce mixed or incomplete cellular profiles — transcripts bleeding across boundaries, or cells whose nuclei are never detected.
TRACER refines cellular representations by leveraging two signals already present in the data: gene–gene coherence (which genes tend to co-occur in real cell types) and spatial co-localization of transcripts. It does this without external annotations or a reference atlas, resolving mixed profiles and reconstructing partial cells to give a more complete and biologically coherent picture of the tissue.
Installation
TRACER is implemented in Python. The public code release is expected alongside peer-reviewed publication; installation instructions and a PyPI/GitHub link will appear here at that time. In the meantime, see the preprint for methodological details, or contact the lab for early access.
Inputs
TRACER expects per-transcript spatial data with, at minimum, a gene label and spatial coordinates, optionally with an initial cell assignment from segmentation.
| Input | Description |
|---|---|
| Transcript table | Per-molecule records: gene, x, y (and z if available) |
| Initial segmentation | Optional cell IDs from a segmentation method, to be refined |
| Cell / nucleus centroids | Optional, used to anchor reconstruction of partial cells |
Quick start
A typical analysis proceeds in three conceptual steps:
# 1. Load imaging-based spatial transcriptomics data import tracer data = tracer.load_transcripts("sample_transcripts.csv") # 2. Run TRACER to refine and reconstruct cellular profiles result = tracer.run(data) # 3. Export refined cell-by-gene profiles for downstream analysis result.to_anndata("tracer_cells.h5ad")
Outputs
TRACER returns refined cell-by-gene profiles — including reconstructed partial cells — in a standard single-cell container (e.g., AnnData) ready for clustering, cell typing, and downstream spatial analysis with tools such as SpaceMarkers.
Citing TRACER
Yuan L, Zheng Y, Zhang S, Beroukhim R, Deshpande A. Reconstructing biologically coherent cellular profiles from imaging-based spatial transcriptomics. bioRxiv (2026). biorxiv.org/10.64898/2026.03.08.710395