6. Centroid-Level Optimal Transport

6. Centroid-Level Optimal Transport#

scBIOT’s centroid workflow projects each batch down to a few thousand centroid representatives, runs OT on the compressed problem, and then interpolates the displacement field back to every cell. This keeps memory bounded while scaling from benchmarking-sized datasets to the 100-million-cell Tahoe experiment.

Choose one of the recipes below to dive into the matching notebook.