Releases · JurgenKriel/SMINT

SMINT 2.0.0 is the first release since the Zenodo archive (v.1.0.0, tag smint). SMINT is described in Kriel et al., Cell Reports Methods 6(2), 101306 (2026).

pip install "git+https://github.com/JurgenKriel/SMINT.git@v2.0.0"

Highlights

Registration

Same-section registration, rebuilt. The new default method, multiround, uses displacement voting within tissue blocks, then an affine round and a smooth thin-plate-spline round. Each round is kept only if it improves the alignment. On a real Xenium section, the held-out median distance to the nearest cell fell from 12.4 to 5.4 µm.
Snapping refinement. Confidently matched nuclei are placed exactly on their Xenium cells. Nuclei that remain are assigned to free cells and flagged with a neighbour_deviation.
Sequential sections: LDDMM starts from your landmarks. Sections with a large offset now converge: in a synthetic test, the median error went from 695 µm to 0.1 µm.
A napari plugin and an HPC job runner cover loading data, pre-registration, landmarks and running jobs locally or through SLURM, with a choice of GPU type.
smint-alignment command line: st-sm, centroid, pretransform, columns, plus --submit and --watch.

Segmentation

Cellpose 3 and Cellpose 4 (Cellpose-SAM) are both supported.
Distributed whole-slide segmentation with overlapping tiles on local, SLURM or existing Dask workers. It includes the R stitching ported to Python (merge, trim, remove, then reconciling cells with nuclei). The new command is smint-segment-distributed.
run_segmentation.py: model and diameter flags now also apply with --distributed, and --no-gpu runs either mode on the CPU.

Documentation

The site has been rebuilt: executed tutorials, concept guides, task guides, a command-line and API reference, a napari guide and a gallery.

Centroid registration defaults to multiround followed by snapping. x_transformed/y_transformed hold the refined positions, and *_smooth holds the transform-only ones. Use --no-refine / refine=False for the old behaviour.
smint-alignment centroid detects the centroid columns automatically, so --source-cols is only needed for unusual names.
run_lddmm_alignment(init_from_landmarks=True) is the default.
python -m smint.segmentation.distributed_seg is replaced by smint-segment-distributed.
st_align_wrapper and the smint-segmentation console script were removed (see the changelog for their replacements).
Python 3.10 or newer is required.