Skip to content

outputs.md

Latest commit

 

History

History
98 lines (65 loc) · 2.23 KB

outputs.md

File metadata and controls

98 lines (65 loc) · 2.23 KB
Raw

Outputs

The C++ branch currently supports two output styles:

  • legacy
  • parquet

Select them with:

./build/baysor run --output-style legacy ...
./build/baysor run --output-style parquet ...

legacy is the default.

For file-by-file definitions, including exact column order, optional fields, and HDF5 / Parquet layout details, see Output Files.

Legacy Output

The legacy bundle mirrors the familiar Baysor output set:

  • segmentation.csv
  • segmentation_cell_stats.csv
  • segmentation_polygons_2d.json
  • segmentation_polygons_3d.json
  • segmentation_counts.loom or segmentation_counts.tsv
  • segmentation_params.dump.toml
  • segmentation_log.log

Optional:

  • diagnostic_report.html
  • segmentation_plot.html

These files are defined in Output Files.

Xenium Compatibility In Legacy Output

For Xenium-origin inputs, legacy automatically adds the extra fields needed by xeniumranger import-segmentation:

  • segmentation.csv
    • includes transcript_id
    • writes is_noise as true / false
  • segmentation_polygons_2d.json
    • uses GeoJSON FeatureCollection
    • includes properties.cell

This is the recommended output style when the result will be handed off to Xenium Ranger / Xenium Explorer.

Parquet Output

The parquet style is an interop-oriented bundle:

  • molecules.parquet
  • cells.parquet
  • cell_boundaries.parquet
  • cell_boundaries_3d.parquet
  • feature_matrix.h5
  • run_params.toml
  • run.log

Optional:

  • diagnostic_report.html
  • segmentation_plot.html

These files are defined in Output Files.

Boundary outputs use GeoParquet-compatible metadata.

The count matrix is written as a 10x-style HDF5 feature-barcode matrix.

Legacy-Only Flags

These remain meaningful only for legacy output:

  • --polygon-format
  • --count-matrix-format

For parquet, they are ignored with a warning.

Choosing Between Styles

Use legacy when:

  • you want the familiar Baysor outputs
  • you want Xenium Ranger compatibility
  • you need GeoJSON polygons

Use parquet when:

  • you want easier downstream analysis in Python / R / DuckDB
  • you want GeoParquet boundaries
  • you want a 10x-style HDF5 matrix bundle