Describe the bug

Humza is running a cohort and will be using the nuclear signal for Pixie analysis. The nuclear signal is not normalized in segmentation_utils.transform_expression_matrix.

The process for size normalization at https://github.com/angelolab/ark-analysis/blob/main/src/ark/segmentation/segmentation_utils.py#L117-L133 only works for whole cell data, since the "cell_size" and "label" columns only contain the whole cell channel expression columns in between, not the corresponding nuclear channel expression columns. Additionally, normalization by whole cell size should only happen for whole cell signal; the function does not support nuclear signal normalization by nuclear area.

Expected behavior

If nuclear signal is run through Pixie, it should be at least size normalized, if not arcsinh normalized on top of that. transform_expression_matrix should account for this so marker_quantification.create_marker_count_matrices doesn't append raw nuclear signal as normalized.