Configuration

Save input and segmentation settings in a TOML file and pass it with -c. Explicit CLI flags override the file:

baysor run -c config.toml -o out --threads 8 molecules.csv

A minimal config.toml for a table with x, y and gene columns is:

[molecules]
min_molecules_per_cell = 30

[segmentation]
scale = 8.0

These values are examples; choose min_molecules_per_cell and scale for your dataset. Output choices such as --output-style and --plot are CLI flags, not config keys. Use the space-separated form -c config.toml or --config config.toml.

Protocol presets

Download a preset separately from the binary, then adjust it for your data:

Preset	Use
xenium.toml	Xenium column names, control-gene filters, min_molecules_per_cell = 50, prior confidence 0.5 and unassigned label UNASSIGNED.
iss.toml	Sparse ISS data.
osm_fish.toml	osm-FISH data.
starmap.toml	STARmap data.

See the Xenium workflow for a download-and-run command, or example_config.toml for a template listing every key with its default.

Config key reference

Missing keys keep their defaults. Use key = value lines in the sections below; quote strings and use true / false for booleans. # starts a comment. Invalid numeric or boolean values cause an error; unknown keys are ignored, so check spelling carefully.

Derived defaults use integer division and m = max(min_molecules_per_cell, 3).

Top-level keys

Place these before any [section].

Key	Type	Default	Meaning
threads	int	0	Worker threads; auto uses OMP_NUM_THREADS, then physical CPU cores.

[molecules] / [data]

Use [molecules] in new configs. [data] is an alias; if both are present, [data] is applied last.

Key	Type	Default	Meaning
x, y, z	string	x, y, z	Coordinate column names; z is optional.
gene	string	gene	Gene-name column.
qv	string	qv	Quality-value column used by min_qv.
force_2d	bool	false	Ignore the z column.
min_molecules_per_gene	int	1	Drop genes with fewer molecules.
exclude_genes	string	empty	Comma-separated gene names or glob patterns (*, ?).
min_molecules_per_cell	int	0	Minimum molecules expected in a real cell; must be set to a positive value.
confidence_nn_id	int	0	Neighbors for noise-confidence estimation; 0 = max(m / 2 + 1, 5).
min_qv	float	-1	Drop molecules below this quality value; negative disables.
x_min, x_max	float	−∞, +∞	Crop in x.
y_min, y_max	float	−∞, +∞	Crop in y.
z_min, z_max	float	−∞, +∞	Crop in z.
min_molecules_per_segment	int	0	Prior segment threshold; 0 = max(m / 4, 2). Prefer this key in [prior].

[segmentation]

Key	Type	Default	Meaning
scale	float	-1	Approximate cell radius; must be positive unless a usable prior supplies it. A positive value disables prior scale estimation.
scale_std	string	"25%"	Scale variation: absolute value or percentage of scale; replaced when scale is estimated from a prior.
cluster_method	string	mrf	mrf, louvain, leiden or none; legacy alias ica_mrf.
n_clusters	int	0	Auto = 4 for mrf, 10 for graph methods; exact count for mrf, merged target for louvain / leiden.
cluster_resolution	float	1.0	Advanced overclustering resolution for graph methods.
cluster_graph_k	int	15	NCV neighbors for graph clustering and NCV UMAPs.
cluster_n_dims	int	20	NCV dimensions used by graph methods.
cluster_basis_sample_size	int	100000	Maximum NCV basis anchors used by graph methods.
prior_segmentation_confidence	float	0.2	Prior confidence in [0, 1]; see Prior segmentation.
iters	int	500	Maximum segmentation iterations.
tol	float	0	Stop when fewer than this fraction of molecules change assignment over 20 consecutive iterations; 0 runs all iterations.
n_cells_init	int	0	Auto = 2 * floor(n_molecules / m), reduced when a prior is available.
nuclei_genes, cyto_genes	string	empty	Not implemented in C++; setting either makes run fail.
estimate_scale_from_centers	bool	true	Compatibility alias for [prior].estimate_scale_from_prior.
unassigned_prior_label	string	0	Compatibility alias for [prior].unassigned_label.

[prior]

These keys describe the prior input. The second CLI positional argument overrides its type and source. [prior] keys take precedence over the compatibility forms above; explicit CLI flags override both.

Key	Type	Default	Meaning
type	string	none	none, column, image or boundary.
path	string	empty	TIFF mask or boundary-table path.
column_name	string	empty	Column name for a column prior, without the : prefix.
unassigned_label	string	0	Column-prior value meaning no assignment.
min_molecules_per_segment	int	0	Discard smaller prior segments; 0 = max(m / 4, 2).
estimate_scale_from_prior	bool	true	Estimate scale and scale variation from retained prior segments.

[plotting]

Key	Type	Default	Meaning
gene_composition_neighborhood	int	0	Spatial neighbors for NCVs; auto = max(n_genes / 10, m, 3).
gene_composition_neigborhood	int	—	Julia-era misspelling, still accepted; the correctly spelled key takes precedence.
min_pixels_per_cell	int	15	Accepted for compatibility; unused by the C++ pipeline.
max_plot_size	int	3000	Longer side of molecule images in HTML reports, in pixels; values below 1 use the default.
max_z_slices	int	10	Maximum z-layers for 3D polygons; larger stacks are binned. Must be ≥ 1.
ncv_method	string	ri	Compatibility values ri, dense, sparse; the C++ pipeline uses random indexing regardless.

A run records config values and its invocation in segmentation_params.dump.toml (or run_params.toml for Parquet output). Keep your original config as well: the dump does not serialize every setting.