Citation
If AssayLoop is useful to you, please cite the paper. If you use the benchmark data or the evaluation metrics, please also cite AssayBench, which supplies both.
AssayLoop
@article{edwards2026biologyloop,
title={Biology-in-the-loop: Amortized Adaptive Hit Discovery in CRISPR Screens},
author={Edwards, Carl and De Brouwer, Edward and Li, Xiner and Lee, Namkyeong and
Hajiramezanali, Ehsan and Biton, Anne and Mostafavi, Sara and Scalia, Gabriele},
journal={arXiv preprint arXiv:2609.11877},
url={https://arxiv.org/abs/2609.11877},
year={2026}
}
AssayBench
The screen corpus and metadata come from AssayBench.
@article{de2026assaybench,
title = {AssayBench: An Assay-Level Virtual Cell Benchmark for LLMs and Agents},
author = {De Brouwer, Edward and Edwards, Carl and Wu, Alexander and
Collier, Jenna and Heimberg, Graham and Li, Xiner and
Subramaniam, Meena and Hajiramezanali, Ehsan and
Richmond, David and H{\"u}tter, Jan-Christian and
Mostafavi, Sara and Scalia, Gabriele},
journal = {arXiv preprint arXiv:2605.10876},
year = {2026}
}
Underlying data
AssayLoop is built on public resources that carry their own attribution requirements. The repository's data sources table is the authoritative list; the short version:
- BioGRID ORCS — the screens themselves, via the Genentech/assaybench dataset.
- Reactome — the pathway hierarchy behind the Effective Pathways metric and the sunbursts. CC BY 4.0.
- MSigDB (Hallmark, C5 GO:BP, C2 CP) — enrichment priors and pathway diversity. Attribution: Broad Institute, MIT, and the Regents of the University of California.
- STRING v11.5, CORUM 4.1, SIGNOR 3.0 — interaction ground truth for the network-recovery analysis. CC BY 4.0 at these pinned versions.
- DepMap — common-essential gene list, used for the %essential column. The AssayLoop results and UMAP annotations use DepMap Public 26Q1, subject to the DepMap Terms and Conditions; they are not covered by AssayLoop's MIT License.
- PRESAGE — gene embeddings used by several baselines.