Reproducing the paper

The experiments/ folder of the repository contains the scripts of the 2D datasets of the paper (Fig. 2): neural W2OT with the squared Euclidean cost, OT-SI and CloneOT on the T-shape and rotated-moons datasets, each with three training seeds.

pip install -e ".[experiments]"
cd experiments/simulations
for task in tshape rotated; do for seed in 0 1 2; do python run.py --task $task --seed $seed; done; done
python figure.py

The default settings of CloneOT are the settings of the paper, shared by the 2D datasets and all single-cell tasks. The Watermelon tutorial shows the complete single-cell workflow, from the download of the public data to the clone-level evaluation of seen and unseen clones. The single-cell datasets of the paper are public: Watermelon (GEO GSE150949), ClonMapper (GSE291678) and ReSisTrace (GSE223003).