Effects of data transformation and model selection on feature importance in microbiome classification data
Problem
Classifying healthy versus diseased individuals from gut microbiome data is difficult due to data sparsity, compositionality, and population-specificity.
Why it matters
Accurate microbiome-based classification is stated as crucial for advancing microbiome-based therapies.
What the AI did
Machine learning analysis of over 8,500 samples from 24 shotgun metagenomic datasets showed healthy/diseased classification is possible with minimal dependence on algorithm or data transformation choice.
Results
Classification achieved with minimal dependence on algorithm or transformation; presence-absence transformations performed comparably to abundance-based; only a small subset of predictors necessary
Impact
Methodological finding guiding ML practice in microbiome research; no therapeutic deployment stated.
Record
| Venue | Microbiome |
|---|---|
| Published | 2025 |
| Work type | article |
| Authors | Zuzanna Karwowska, Oliver Aasmets, Mait Metspalu, Andres Metspalu, Lili A. Milani, Tõnu Esko, Tomasz Kościółek, Elin Org |
| Citations | 37 (OpenAlex, at import time) |
| References | 43 |
| Model or method | not stated in source |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1186/s40168-024-01996-6 |
| Open access | gold |
| Keywords | microbiome, machine learning, classification, data transformation, metagenomics |
Importance rating
real but narrower advance
Abstract says accurate microbiome-based classification is crucial for advancing microbiome-based therapies; the contribution is methodological (robustness to transformations), with no scale or deployment stated.