How it Works

Data Structure

Datasets are stored as AnnData objects, where observations are cells, variables are genes, and the X is the count matrix. Bento additionally stores molecular coordinates in uns['points'] and polygons as columns in obs.

AnnData adapted to hold spatial data

Shapes

Currently, shapes are stored as GeoSeries columns according to which cell they belong to. These columns are identified with the suffix "_shape". Each element in the GeoSeries is either a shapely Polygon or MultiPolygon. Shape properties are also stored as columns and identified with the corresponding shape name as the prefix e.g. "cell", "nucleus", etc.

Points

For fast spatial queries, Bento indexes points to shape layers upfront, and saves them as columns points, denoted as "shape index" above. For example, "cell" and "nucleus" columns are added to indicate whether points are within the shape.

Metadata for points are stored as matrices uns. These metadata matrices are the same length as points, which makes it easy to query points and associated metadata. All metadata keys are registered to uns['points_metadata'], which is used to keep them in sync.

RNAflux

RNAflux is a method for quantifying spatial composition gradients in the cell.

RNAforest input features

The following describes input features of the RNAforest model.

Categories	Features
Distance	Cell inner proximity: The average distance between all points within the cell to the cell boundary normalized by cell radius. Values closer to 0 denote farther from the cell boundary, values closer to 1 denote closer to the cell boundary. Nucleus inner proximity: The average distance between all points within the nucleus to the nucleus boundary normalized by cell radius. Values closer to 0 denote farther from the nucleus boundary, values closer to 1 denote closer to the nucleus boundary. Nucleus outer proximity: The average distance between all points within the cell and outside the nucleus to the nucleus boundary normalized by cell radius. Values closer to 0 denote farther from the nucleus boundary, values closer to 1 denote closer to the nucleus boundary.
Symmetry	Cell inner asymmetry: The offset between the centroid of all points within the cell to the centroid of the cell boundary, normalized by cell radius. Values closer to 0 denote symmetry, values closer to 1 denote asymmetry. Nucleus inner asymmetry: The offset between the centroid of all points within the nucleus to the centroid of the nucleus boundary, normalized by cell radius. Values closer to 0 denote symmetry, values closer to 1 denote asymmetry. Nucleus outer asymmetry: The offset between the centroid of all points within the cell and outside the nucleus to the centroid of the nucleus boundary, normalized by cell radius. Values closer to 0 denote symmetry, values closer to 1 denote asymmetry.
Dispersion	Point dispersion: The second moment of all points in a cell relative to the centroid of the total RNA signal. This value is normalized by the second moment of a uniform distribution within the cell boundary. Nucleus dispersion: The second moment of all points in a cell relative to the centroid of the nucleus boundary. This value is normalized by the second moment of a uniform distribution within the cell boundary.
Density	L-function max: The max value of the L-function evaluated at r=[1,d], where d is half the cell’s maximum diameter. L-function max gradient: The max value of the gradient of the above L-function. L-function min gradient: The min value of the gradient of the above L-function. L monotony: The correlation of the L-function and r=[1,d]. L-function at d/2: The value of the L-function evaluated at ¼ of the maximum cell diameter. The L-function measures spatial clustering of a point pattern over an area of interest.