3D polygons or 2D polygons as segmentation

[ruqianl]

Hi @dcjones ,

Just wondering if you have tested/compared, whether using 3D polygons (3D segmentation) is better than using the 2D projection version for downstream analysis.

Thanks, RL

[dcjones]

Hi RL,

I think it may depend on what kind of downstream analysis you're doing. I use the 2D consensus polygons for visualizations, since it's more readable, and also for any downstream tools that can't easily be made to work with 3D segmentation. The layered 3D polygons are what proseg is actually using to estimate transcript counts, so ideally any quantitative analysis should probably be based on that.

[ruqianl]

Thanks for your reply!

The layered 3D polygons are what proseg is actually using to estimate transcript counts, so ideally any quantitative analysis should probably be based on that.

Is there a quick way of obtaining the raw counts per 3D seg polygon from the output files?

[dcjones]

To clarify, the counts that proseg reports (in the expected-counts file) are always with respect to the 3d cells, which is what's being used internally. The flattened 2D polygons are meant as an approximation for better visualization.

[ruqianl]

Thanks for that.

It feels like the raw counts in the 3d cells should be helpful too.