MGM2 as a Multimodal Foundation Model for Microbial Community States and Responses
- 1Huazhong University of Science and Technology
Abstract
Microbial communities sustain ecosystems and influence host health, yet connecting their molecular diversity to responses under intervention remains difficult. Here we introduce MGM2, a multimodal foundation model integrating DNA language-model representations of microbial sequences, quantitative abundances and sample descriptions. After pretraining on 1.8 million microbiomes, its frozen representations achieved a mean macro-AUROC of 0.91 on later MGnify projects absent from pretraining. These representations also supported fecal microbiota transplantation outcome prediction, recovering 68% of observed high-change taxa. Adding molecular structures enabled prediction of drug effects in nested five-fold evaluation, while measured perturbed states informed subsequent recovery. To interpret the learned representations, sparse decomposition resolved taxonomic, ecological and abundance-related features. Together, these results show how a shared representation of molecular identity and community context connects large-scale microbiome learning to the prediction and interpretation of community change.
Related
- Microbial community responses to ceramic amendment in coastal sediments of a seasonally hypoxic bay
- From osmolytes to oxidative stress: Temporal exometabolite shifts and microbial community responses in rehydrated biocrusts
- Community Assembly Mechanisms Facilitate Understanding of Microbial Responses to Habitat Transition along a Spatial Environmental Continuum in Alpine Plateaus
- Long-read shotgun metagenomics identifies distinct urogenital microbial community states associated with functional potential and urinary tract infection during pregnancy
- Cube-based screening identifies a quinoa-derived synthetic microbial community that promotes plant growth and modulates root epidermal responses under salt stress
BibTeX
@misc{zhang2026mgm2,
title = {{MGM2 as a Multimodal Foundation Model for Microbial Community States and Responses}},
author = {H. Zhang and Y. Zhang and Y. Qi and T. Liu and L. Zhao and R. Yang and K. Ning},
howpublished = {bioRxiv},
year = {2026},
month = jul,
doi = {10.64898/2026.07.20.739063},
url = {https://www.biorxiv.org/content/10.64898/2026.07.20.739063v2},
note = {CC BY-NC},
}
RIS
TY - UNPB
TI - MGM2 as a Multimodal Foundation Model for Microbial Community States and Responses
AU - H. Zhang
AU - Y. Zhang
AU - Y. Qi
AU - T. Liu
AU - L. Zhao
AU - R. Yang
AU - K. Ning
PY - 2026
DA - 2026/07/20
PB - bioRxiv
DO - 10.64898/2026.07.20.739063
UR - https://www.biorxiv.org/content/10.64898/2026.07.20.739063v2
AB - Microbial communities sustain ecosystems and influence host health, yet connecting their molecular diversity to responses under intervention remains difficult. Here we introduce MGM2, a multimodal foundation model integrating DNA language-model representations of microbial sequences, quantitative abundances and sample descriptions. After pretraining on 1.8 million microbiomes, its frozen representations achieved a mean macro-AUROC of 0.91 on later MGnify projects absent from pretraining. These representations also supported fecal microbiota transplantation outcome prediction, recovering 68% of observed high-change taxa. Adding molecular structures enabled prediction of drug effects in nested five-fold evaluation, while measured perturbed states informed subsequent recovery. To interpret the learned representations, sparse decomposition resolved taxonomic, ecological and abundance-related features. Together, these results show how a shared representation of molecular identity and community context connects large-scale microbiome learning to the prediction and interpretation of community change.
ER -