Best practices for developing microbiome-based disease diagnostic classifiers through machine learning
Problem
Microbiome-based ML disease diagnostic models lack a standardized workflow; tool and parameter choices hurt performance and generalizability.
Why it matters
Unreliable diagnostic models hinder microbiome-based disease detection and clinical adoption.
What the AI did
ML pipeline steps (preprocessing, batch effect removal, algorithms) were sequentially optimized and benchmarked across 156 tool-parameter-algorithm combinations on 83 cohorts.
Results
156 combinations benchmarked on 83 cohorts across 20 diseases; four preprocessing methods best for regression-type algorithms; ComBat (sva R package) best for batch effect removal.
Impact
Delivers a benchmarked, generally applicable workflow for microbiome ML diagnostics.
Record
| Venue | Gut Microbes |
|---|---|
| Published | 2025 |
| Work type | article |
| Authors | P Li, Min Li, Wei‐Hua Chen |
| Citations | 33 (OpenAlex, at import time) |
| References | 76 |
| Model or method | not stated in source |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1080/19490976.2025.2489074 |
| Open access | gold |
| Keywords | gut microbiome, machine learning, diagnostics, benchmarking, batch effects, AUC |
Importance rating
significant — thousands to millions affected
Abstract claims a generally applicable workflow for diagnostic models across a range of diseases, benchmarked on 83 cohorts across 20 diseases, potentially aiding future medical diagnostics; broad but methodological impact.