Machine learning approaches in microbiome research: challenges and best practices
Problem
Microbiome ML workflows suffer domain-specific pitfalls in preprocessing, feature selection, modeling, and interpretation, risking unreliable disease diagnosis and biomarker discovery.
Why it matters
Misleading microbiome-based diagnoses or biomarkers could misdirect clinical research and treatment.
What the AI did
Compared ML pipeline approaches on a multi-cohort colorectal cancer shotgun metagenomics dataset and issued best-practice recommendations from COST Action ML4Microbiome.
Results
Multivariate feature selection (Statistically Equivalent Signatures) reduced classification error; compositional transformations and filtering did not always improve performance (no numbers given).
Impact
Provides recommendations to make ML-based microbiome disease diagnosis and biomarker discovery more reliable.
Record
| Venue | Frontiers in Microbiology |
|---|---|
| Published | 2023 |
| Work type | article |
| Authors | Γεώργιος Παπουτσόγλου, Sonia Tarazona, Marta Belchior Lopes, Thomas Klammsteiner, Eliana Ibrahimi, Julia Eckenberger, Pierfrancesco Novielli, Alberto Paolo Tonda et al. |
| Citations | 137 (OpenAlex, at import time) |
| References | 115 |
| Model or method | Statistically Equivalent Signatures algorithm among ML approaches |
| Organization | COST Action ML4Microbiome |
| Primary source | https://doi.org/10.3389/fmicb.2023.1261889 |
| Open access | gold |
| Keywords | microbiome, machine learning, best practices, colorectal cancer, biomarker discovery |
Importance rating
incremental / niche
Best-practice recommendations for microbiome ML pipelines improve reliability of diagnosis/biomarker discovery; methodological guidance with narrow, indirect impact.