https://premier-microbiome.org/wp-content/uploads/sites/6/2026/05/A-foundation-model-for-microbial-growth-dynamics-.pdf
This work presents a large-scale foundation model designed to learn generalizable representations of microbial growth dynamics across diverse species, environmental conditions, and community contexts. The model is trained using self-supervised learning on approximately 370,000 experimental and simulated microbial growth curves, enabling it to capture essential dynamical features in compact, low-dimensional latent embeddings.
These learned representations allow accurate reconstruction of raw growth data and significantly enhance performance across a range of downstream tasks. The authors demonstrate few-shot learning for antibiotic classification and concentration prediction, improved forecasting of both simulated and experimental microbial communities, and inference of total population abundance from relative-abundance measurements. By extracting transferable structure from heterogeneous growth datasets, this model provides a unifying framework for analyzing and predicting microbial population behavior in settings where task-specific data are limited, with implications for antibiotic testing, microbiome engineering, and systems biology.
This article is a preprint and has not yet been certified by peer review.
You, L., David, L., Holmes, Z. A., Shyti, I., Hoffman, A. L., Duncker, K. E., Ma, H. R., Zhou, Z., Lee, D., Maddamsetti, R., Kim, K., Şimşek, E., Hamrick, G. S., Son, H., Villalobos, C. A., Lu, J., Ha, Y., Shende, A. R., Yao, Z., Liu, S., Shapiro, D. M., & Kholina, K. (2025). A foundation model for microbial growth dynamics. https://doi.org/10.64898/2025.12.01.691707
