Outputs

From segger segment

segger_segmentation.parquet is the main output, with the following columns:

• row_index — index into the input transcripts

• segger_cell_id — the assigned cell

• segger_similarity, similarity_threshold — the assignment score and its per-gene cutoff

• converged — whether the cutoff was computed directly for that gene or backfilled from the global median

• x, y, feature_name

• filtered — assigned, converged, and above threshold; the recommended keep mask

Segger also writes segger_anndata.h5ad, a cell x gene AnnData table, unless --save-anndata is disabled (this default will be deprecated soon — use segger export anndata instead). With --debug, it additionally writes a debug/ directory with run params, tiles, predictions, and the trainer checkpoint.

import polars as pl

seg = pl.read_parquet("outputs/segger_segmentation.parquet")
assigned = seg.filter(pl.col("filtered"))

From segger export

segger export anndata \
    -s /path/to/save/outputs/segger_segmentation.parquet \
    -o /path/to/export/

Element	Output file	Description	Parameters
anndata	adata.h5ad	Cell x gene AnnData table with spatial coordinates	--min-counts
boundaries	cell_boundaries.parquet	One polygon per cell, as GeoParquet	--method, --chaikin-iterations
transcripts	transcripts.parquet	Per-transcript table with assigned cell IDs	--include-all-transcripts
spatialdata	<sdata>.zarr (in place)	Adds segger’s outputs into an existing SpatialData store	--sdata, --sdata-transcripts-name, --sdata-cell-boundaries-name, --sdata-table-name

See How To for how to build the SpatialData store in the first place, and segger export for a full description of every parameter.