GENESIS results
Every score behind the figures of Single-cell perturbation prediction near experimental reproducibility with GENESIS: GENESIS, the baselines, the comparators (STATE, CellFlow, PerturbDiff) and the split-half references, per condition. The code is at github.com/rickconci/GENESIS, and the trained models at RConci/GENESIS.
Rebuild the figures' tables
Inside a clone of the code (uv sync), download the scores and rebuild every table the figures are drawn from:
uvx --from huggingface_hub hf download RConci/GENESIS_results --repo-type dataset --include 'core_results/*' --local-dir .
uv run bash scripts/figures/extract_all.sh
extract_all.sh rebuilds core_deliverables/data/ from these scores and checks every table byte for byte. To
recompute the scores themselves from the models, see the code's scripts/figures/README.md (level 3).
What is here
core_results/figureN/ holds the scores of figure N, in the folders its scripts/figures/figureN/score/ drivers write.
core_results/MANIFEST.tsv lists every file with its size, sha256 and the driver that made it.
figure4/progeny/eval_cache_n3072/ is an input rather than a score: the measured cells of Figure 4's test conditions
and their controls, on its 3,072 genes.
Licence
Each file carries the licence of the data it was computed from (LICENSE.md):
| Files | Computed from | Licence |
|---|---|---|
every file with parse or pbmc in its path |
the Parse PBMC cytokine atlas, CC BY-NC 4.0 | CC BY-NC 4.0: non-commercial use, with credit |
| every other file | Tahoe-100M, CC0 1.0 | CC BY 4.0: any use, with credit |
Credit means citing the GENESIS paper, plus the dataset a file was computed from.
Data
- Zhang J. et al. Tahoe-100M: A Giga-Scale Single-Cell Perturbation Atlas for Context-Dependent Gene Function and Cellular Modeling. bioRxiv (2025). doi:10.1101/2025.02.20.639398
- Parse Biosciences. 10 Million Human PBMCs in a Single Experiment. parsebiosciences.com/datasets/10-million-human-pbmcs-in-a-single-experiment
- Downloads last month
- 23