Self-supervised learning enables robust microbiome predictions in data-limited and cross-cohort settings
Authors:
- Liron ZahaviLZLiron Zahavi
- Zachary LevineZLZachary Levine
- Anastasia GodnevaAGAnastasia Godneva
- Veronika DubinkinaVDVeronika Dubinkina
- Raja DhirRDRaja Dhir
- Katherine S. PollardKPKatherine S. Pollard
- Adina WeinbergerAWAdina Weinberger
- Eran SegalESEran Segal
Preprint
Abstract
The gut microbiome plays a crucial role in human health, but machine learning applications in this field face significant challenges, including limited labeled data availability, high dimensionality, and batch effects across different cohorts. To address these limitations, we developed representation learning models for gut microbiome metagenomic data, drawing inspiration from foundation models approaches based on self-supervised and transfer learning principles. By leveraging a large collection of 85,364 metagenomic samples, we implemented multiple self-supervised learning methods, including masked autoencoders with varying masking rates and adapted single-cell RNA sequencing models (scVI and scGPT), to generate embeddings from bacterial abundance profiles. These learned representations demonstrated significant advantages over raw bacterial abundances in two key scenarios: first, when training predictive models with very limited labeled data, improving prediction performance for age (r = 0.14 vs. 0.06), BMI (r = 0.16 vs. 0.11), visceral fat mass (r = 0.25 vs. 0.18), and drug usage classification (PR-AUC = 0.81 vs. 0.73); and second, when generalizing predictions across different cohorts, consistently outperforming models based on raw abundances in cross-dataset evaluation. Our approach provides a valuable framework for leveraging self-supervised representation learning to overcome the data limitations inherent in microbiome research, potentially enabling more robust and generalizable machine learning applications in this field.Keywords
Funding Information
- Crown Human Genome Center
- the Larson Charitable Foundation New Scientist Fund
- the Else Kröner Fresenius Foundation
- the White Rose International Foundation
- the Ben B. and Joyce E. Eisenberg Foundation
- the Nissenbaum Family
- Marcos Pinheiro de Andrade and Vanessa Buchheim
- Lady Michelle Michels
- Aliza Moussaieff
- the Minerva Foundation
- the Federal German Ministry for Education and Research
- the European Research Council
- he Israel Science Foundation
This publication has 17 references indexed in Scilit:
- Deep phenotyping of health–disease continuum in the Human Phenotype ProjectNature Medicine, 2025
- scGPT: toward building a foundation model for single-cell multi-omics using generative AINature Methods, 2024
- Bacterial SNPs in the human gut microbiome associate with host BMINature Medicine, 2023
- Transfer learning enables predictions in network biologyNature, 2023
- DeepGeni: deep generalized interpretable autoencoder elucidates gut microbiota for better cancer immunotherapyScientific Reports, 2023
- An expanded reference map of the human gut microbiome reveals hundreds of previously unknown speciesNature Communications, 2022
- Environmental factors shaping the gut microbiome in a Dutch populationNature, 2022
- Highly accurate protein structure prediction with AlphaFoldNature, 2021
- DeepMicro: deep representation learning for disease prediction based on microbiome dataScientific Reports, 2020
- SciPy 1.0: fundamental algorithms for scientific computing in PythonNature Methods, 2020