Fixed a problem with huge integers in cell ids

README.md

@@ -135,7 +135,7 @@ In some cases, you may want to use another segmentation as a prior for Baysor. T
 baysor run [ARGS] MOLECULES_CSV [PRIOR_SEGMENTATION]
 ```
 
-Here, `PRIOR_SEGMENTATION` can be a path to a binary image with a segmentation mask, an image with integer cell segmentation labels or a column name in the `MOLECULES_CSV` with integer cell assignment per molecule. In the latter case, the column name must have `:` prefix, e.g. for column `cell` you should use `baysor run [ARGS] molecules.csv :cell`. In case the image is too big to be stored in the tiff format, Baysor supports MATLAB '.mat' format: it should contain a single field with an integer matrix for either a binary mask or segmentation labels. When loading the segmentation, Baysor filters segments that have less than `min-molecules-per-segment` molecules. It can be set in the toml config, and the default value is `min-molecules-per-segment = min-molecules-per-cell / 4`. **Note:** only CSV column prior is currently supported for 3D segmentation.
+Here, `PRIOR_SEGMENTATION` can be a path to a binary image with a segmentation mask, an image with integer cell segmentation labels or a column name in the `MOLECULES_CSV` with integer cell assignment per molecule (`0` value means no assignment). In the latter case, the column name must have `:` prefix, e.g. for column `cell` you should use `baysor run [ARGS] molecules.csv :cell`. In case the image is too big to be stored in the tiff format, Baysor supports MATLAB '.mat' format: it should contain a single field with an integer matrix for either a binary mask or segmentation labels. When loading the segmentation, Baysor filters segments that have less than `min-molecules-per-segment` molecules. It can be set in the toml config, and the default value is `min-molecules-per-segment = min-molecules-per-cell / 4`. **Note:** only CSV column prior is currently supported for 3D segmentation.
 
 To specify the expected quality of the prior segmentation you may use `prior-segmentation-confidence` parameter. The value `0.0` makes the algorithm ignore the prior, while the value `1.0` restricts the algorithm from contradicting the prior. Prior segmentation is mainly needed for the cases where gene expression signal is not enough, e.g. with very sparse protocols (such as ISS or DARTFISH). Another potential use case is high-quality data with a visible sub-cellular structure. In these situations, setting `prior-segmentation-confidence > 0.7` is recommended. Otherwise, the default value `0.2` should work well.
 

src/data_loading/cli_wrappers.jl

@@ -2,6 +2,7 @@ import HDF5
 import JSON
 import LinearAlgebra: Adjoint
 using ProgressMeter
+using StatsBase: denserank
 
 Polygons = Union{Dict{Int, Matrix{T}}, Dict{String, Dict{Int, Matrix{T}}}} where T <: Real
 
@@ -15,6 +16,7 @@ function parse_prior_assignment(pos_data::Matrix{Float64}, prior_segmentation::V
         error("The prior segmentation column '$col_name' must not contain negative numbers")
     end
 
+    prior_segmentation = denserank(prior_segmentation) .- 1
     filter_segmentation_labels!(prior_segmentation, min_molecules_per_segment=min_molecules_per_segment)
     scale, scale_std = estimate_scale_from_assignment(pos_data, prior_segmentation; min_mols_per_cell=min_mols_per_cell)