Citation

If you use scMultiBench (or the multibench Python API) in your research, please cite the benchmark paper. When you run one of the wrapped integration methods, also cite that method's own publication.

TL;DR

Cite Liu, Ding et al., Nature Methods (2025) for the benchmark, and cite each integration method you actually run — mtb.method_info(name) points you at every tool's reference and code.

---

The paper

Liu C, Ding S, Kim HJ, Long S, Xiao D, Ghazanfar S, Yang P. Multitask benchmarking of single-cell multimodal omics integration methods. Nature Methods, 22, 2449–2460 (2025). https://doi.org/10.1038/s41592-025-02856-3

BibTeX

scmultibench.bib

@article{liu2025scmultibench,
  title   = {Multitask benchmarking of single-cell multimodal omics integration methods},
  author  = {Liu, Chunlei and Ding, Sichang and Kim, Hani Jieun and
             Long, Siqi and Xiao, Di and Ghazanfar, Shila and Yang, Pengyi},
  journal = {Nature Methods},
  volume  = {22},
  pages   = {2449--2460},
  year    = {2025},
  doi     = {10.1038/s41592-025-02856-3},
  url     = {https://doi.org/10.1038/s41592-025-02856-3},
  publisher = {Nature Publishing Group}
}

---

Cite the methods you run

Every method you run through multibench is third-party software with its own paper - cite it alongside scMultiBench whenever its output appears in your manuscript. mtb.method_info(name) returns each method's metadata, including pointers to its upstream repository and reference.

---

Source code

The benchmark code, the wrapped method commands, and this multibench API live in the lab repository:

• GitHub: https://github.com/PYangLab/scMultiBench

software.bib

@software{scmultibench_code,
  author    = {Liu, Chunlei and Ding, Sichang and Yang, Pengyi},
  title     = {{scMultiBench}: a benchmark and Python API for single-cell
               multimodal omics integration},
  url       = {https://github.com/PYangLab/scMultiBench},
  note      = {multibench Python package},
  year      = {2025}
}

Reproducibility

For methods reporting, mtb.run(...) returns a RunResult whose .cmd field is the exact command line that was executed. Recording it alongside the method's conda environment and the scMultiBench version (multibench.__version__) makes a run fully reproducible.