Microbiome Preprocessing Machine Learning Pipeline
Problem
Raw 16S microbiome feature counts may be a suboptimal representation for machine learning classification tasks.
Why it matters
Preprocessing choices affect the reliability of microbiome-based machine learning conclusions in research and clinics.
What the AI did
The pipeline tested combinations of preprocessing steps (log transformation, taxonomy-level feature merging, z-scoring) and measured each step's contribution to classification AUC.
Results
Log-transformed counts more informative than relative counts; taxonomy-level merging improved AUC; z-scoring had very limited effect.
Impact
Findings show preprocessing materially affects microbiome ML classification accuracy, though no numeric values are given.
Record
| Venue | Frontiers in Immunology |
|---|---|
| Published | 2021 |
| Work type | article |
| Authors | Yoel Y Jasner, Anna Belogolovski, Meirav Ben Izhak, Omry Koren, Yoram Louzoun |
| Citations | 31 (OpenAlex, at import time) |
| References | 38 |
| Model or method | not stated in source |
| Organization | not stated in source |
| Primary source | https://doi.org/10.3389/fimmu.2021.677870 |
| Open access | gold |
| Keywords | microbiome, 16S sequencing, machine learning, preprocessing, AUC |
Importance rating
incremental / niche
Methodological preprocessing study for 16S microbiome ML classification; improves AUC but no disease, patient, or application scale stated. Incremental methodological advance.