RavshanjonEminov/microbiome-gi-cancer-benchmark

Microbiome ML Benchmark · GI Cancer Diagnosis

  ![5 cohorts]() ![5 models]()

A leak-free, reproducible benchmark of five classical ML model families for microbiome-based GI cancer diagnosis across five public cohorts. Every metric is computed from real model runs on real public data — no synthetic taxa, no fabricated numbers.

Full source & docs: github.com/RavshanjonEminov/microbiome-gi-cancer-benchmark

Model Leaderboard

AUROC mean ± SD, 5×5 repeated stratified CV across four shotgun WMS cohorts.

AUROC by Cohort

References

Primary datasets (re-run in this benchmark)

[1] Zeller G et al. Potential of fecal microbiota for early-stage detection of colorectal cancer. Mol Syst Biol 10:766, 2014. https://doi.org/10.15252/msb.20145645

[2] Feng Q et al. Gut microbiome development along the colorectal adenoma–carcinoma sequence. Nat Commun 6:6528, 2015. https://doi.org/10.1038/ncomms7528

[3] Yu J et al. Metagenomic analysis of faecal microbiome as a tool towards targeted non-invasive biomarkers for colorectal cancer. Gut 66(1):70–78, 2017. https://doi.org/10.1136/gutjnl-2015-309800

[4] Vogtmann E et al. Colorectal cancer and the human gut microbiome: reproducibility with whole-genome shotgun sequencing. PLOS ONE 11(3):e0151467, 2016. https://doi.org/10.1371/journal.pone.0151467

[5] Kostic AD et al. Genomic analysis identifies association of Fusobacterium with colorectal carcinoma. Genome Res 22(2):292–298, 2012. https://doi.org/10.1101/gr.126573.111

Meta-analysis reference baseline

[6] Wirbel J et al. Meta-analysis of fecal metagenomes reveals global microbial signatures specific for colorectal cancer. Nature Medicine 25:679–689, 2019. https://doi.org/10.1038/s41591-019-0458-7

Cross-cancer literature figures (reported, not re-run)

[7] Nagata N et al. Metagenomic identification of microbial signatures predicting pancreatic cancer. Gastroenterology 163(1):222–238, 2022. https://doi.org/10.1053/j.gastro.2022.03.054

[8] Kartal E et al. A faecal microbiota signature with high specificity for pancreatic cancer. Gut 71(7):1359–1372, 2022. https://doi.org/10.1136/gutjnl-2021-324755

[9] Half E et al. Fecal microbiome signatures of pancreatic ductal adenocarcinoma patients. Sci Rep 9:16801, 2019. https://doi.org/10.1038/s41598-019-53041-4

[10] Zhang J et al. Using machine learning to identify the gut microbiome associated with gastric cancer. Appl Microbiol Biotechnol 105:4045–4054, 2021. https://doi.org/10.1007/s00253-020-11043-7

[11] Wu H et al. Gut microbiome signatures in gastric cancer: a multi-cohort meta-analysis. Front Cell Infect Microbiol 14:1397466, 2024. https://doi.org/10.3389/fcimb.2024.1397466

ML software

[12] Pedregosa F et al. Scikit-learn: Machine learning in Python. JMLR 12:2825–2830, 2011.

[13] Chen T & Guestrin C. XGBoost: A scalable tree boosting system. KDD 2016. https://doi.org/10.1145/2939672.2939785

[14] Ke G et al. LightGBM: A highly efficient gradient boosting decision tree. NeurIPS 2017.

[15] Lundberg SM & Lee S-I. A unified approach to interpreting model predictions (SHAP). NeurIPS 2017.

Acknowledgments

The cohort data used in this benchmark were collected and shared by the original study authors — Zeller, Feng, Yu, Vogtmann, and Kostic — whose open-data practices make reproducible microbiome ML research possible. Processed abundance tables are redistributed via DAIZHENWEI/meta-analysis-microbiome and knights-lab/MLRepo. The within-study AUROC reference matrix is from zellerlab/crc_meta (Wirbel et al. 2019).

Built with scikit-learn, XGBoost, LightGBM, Streamlit, and Plotly.

License

Code: MIT. Datasets retain their original licences.