DeepMicro: deep representation learning for disease prediction based on microbiome data
Problem
Predicting human diseases from microbiome data is difficult because the data is extremely high-dimensional (hundreds of thousands of features) while sample sizes are small, making learned models sparse and weak.
Why it matters
Reliable disease prediction from microbiome profiles could enable earlier detection and better health outcomes.
What the AI did
DeepMicro uses autoencoders to learn a robust low-dimensional representation of high-dimensional microbiome data, then applies machine learning classification algorithms on that representation for disease prediction.
Results
Abstract states DeepMicro outperforms current methods; specific metrics not stated in source (abstract truncated)
Impact
Provides a peer-reviewed deep learning framework that reportedly improves microbiome-based disease prediction over existing methods.
Record
| Venue | Scientific Reports |
|---|---|
| Published | 2020 |
| Work type | article |
| Authors | Min Oh, Liqing Zhang |
| Citations | 189 (OpenAlex, at import time) |
| References | 28 |
| Model or method | autoencoders (DeepMicro framework) |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1038/s41598-020-63159-5 |
| Open access | gold |
| Keywords | microbiome, deep learning, autoencoders, disease prediction, representation learning |
Importance rating
real but narrower advance
Microbiome-based disease prediction could enable earlier detection and better health outcomes; validated on five datasets but abstract states no scale of clinical impact.