Skip to content

Segmentation-free analysis

Analyze local gene composition without assigning molecules to cells:

baysor segfree -m 30 -k 100 -o ncvs.loom molecules.csv

Baysor computes one neighborhood composition vector (NCV) per molecule from its nearest neighbors, then log-transforms it. -k / --k-neighbors sets the neighborhood size: larger values emphasize broader spatial patterns. If omitted, it is max(n_genes / 10, min_molecules_per_cell, 3) (integer division).

-m is required (CLI or config) for noise-estimation defaults even when -k is set. No scale or prior is needed. CSV, Parquet and Xenium manifests are accepted; use -c config.toml for column mappings and filters. See Input data and Configuration.

Output

-o names a Loom file (default ncvs.loom), not a directory:

  • /matrix — genes × molecules, one log-transformed NCV per column
  • /col_attrs/Name — molecule names V1 … VN
  • /col_attrs/ncv_color — local gene-composition color
  • /col_attrs/confidence — molecule confidence from the noise model

These are overlapping neighborhoods, not segmented cells or a cell count matrix. --threads works as for run. See the CLI reference or baysor segfree --help for all options.