Define and benchmark challenges in single-cell data science via open standards and continuous leaderboards to promote method development and guide users in method selection. This organization hosts the code, tasks, datasets, and docs behind the Open Problems platform.
- Website & docs: openproblems.bio · Documentation
- Leaderboards: Benchmarks
- Datasets browser: Datasets
- Community: Discord · Interest form
- Core platform -
openproblems
: the living, extensible, community‑guided benchmarking framework. - Common datasets -
datasets
: workflows for managing and processing common datasets. - Benchmarks as repositories: task‑specific repos named
task_*
(e.g.,task_batch_integration
,task_label_projection
,task_spatially_variable_genes
, etc.). - Task template -
task_template
: scaffolding to start a new benchmark with the correct structure. - Shared libraries & images -
core
: helper R/Python packages and base Docker images used across tasks.
More detailed documentation below.
- Tasks define an API, reference datasets, and quantitative metrics.
- Methods implement that task API.
- Continuous evaluation runs standardized workflows to score methods and update leaderboards.
- Reproducibility is enforced via containers and declarative workflows.
Tech stack highlights:
- Nextflow for portable workflows; Viash for modular components; AnnData for standardized I/O; GitHub Actions for CI; deployment via Nextflow Tower with cloud backends (e.g., AWS Batch/S3). See the Technology stack for details.
-
Explore current leaderboards Check out live tasks and results on the Benchmarks page.
-
Run a benchmark locally Read the platform Documentation for install requirements and common commands. Components are containerized; workflows run on laptop, HPC, or cloud.
-
Add your method to a task Follow the docs ("Create component → Add a method") and open a PR in the corresponding
task_*
repo. See repo READMEs for task‑specific APIs. -
Propose or start a new task Start from
task_template
and the docs ("Create a new task"). Open an issue to coordinate scope and maintainership. -
Join the community
- Chat: Discord
- Working meetings: Join our weekly working meetings
- Announcements & events: Interest form / mailing list
- Events archive: openproblems.bio/events
- Governance: Open, consensus‑seeking model with defined roles (Core team, Task teams, Infrastructure, etc.). Read the current Governance.
- Code of Conduct: We follow the Contributor Covenant. Report issues to [email protected]. See the full Code of Conduct.
If you use Open Problems, please cite:
Luecken, M.D., Gigante, S., Burkhardt, D.B. et al. Defining and benchmarking open problems in single‑cell analysis. Nature Biotechnology (2025). https://doi.org/10.1038/s41587-025-02694-w
To reference specific tasks or datasets, please cite the corresponding task or dataset publications mentioned in the task descriptions and dataset pages. For example, to reference our Open Problems multimodal BMMC datasets, please cite Luecken et al., NeurIPS 2021.
Also see earlier NeurIPS challenge reports and proceedings referenced on the Events page.
Open Problems is free open source software and distributed under the MIT License. However, Open Problems tasks may include references to data or code distributed by a third party under a different license. If any question about license arises, please consult the specific repository in which a particular asset is hosted. If further clarification is needed, please open a GitHub Issue referencing the asset if any clarification is needed.
- Code in this org is MIT unless stated otherwise in the repo.
- The website repo uses mixed licensing: Markdown/JSON content under CC‑BY‑4.0 and code under MIT (see that repo's LICENSE files).
- Datasets retain their original source licenses/terms; check dataset pages before downstream use.
Open Problems is supported by a growing community and sponsors including the Chan Zuckerberg Initiative, Data Intuitive, Helmholtz Munich, Saturn Cloud, and Seqera. See the website for the latest list.
- Primary languages: Python and R (task repos may include Bash/Nextflow/TeX for workflows and docs).
- CI builds and unit tests run via GitHub Actions; component images are maintained centrally in
core
.
- Issues and PRs: use the relevant repo (
openproblems
ortask_*
). - Community and governance questions: [email protected].
- Real‑time chat: Discord.