A foundation model for microbial growth dynamics.

Area of Science:

• Microbiology
• Computational Biology
• Systems Biology

Background:

• Microbial growth dynamics offer valuable insights for applications like antibiotic testing and microbiome engineering.
• High dimensionality of growth data and limited datasets hinder generalizable modeling.
• Existing methods struggle with diverse microbial systems and contexts.

Purpose of the Study:

• To develop a foundation model for microbial growth dynamics.
• To learn transferable, low-dimensional representations from diverse growth data.
• To enhance predictive performance in downstream microbial analysis applications.

Main Methods:

• Trained a large-scale, self-supervised representation model on approximately 370,000 experimental and simulated microbial growth curves.
• Utilized diverse microbial species, environmental conditions, and community contexts for training.
• Learned latent embeddings to capture essential dynamical features and enable data reconstruction.

Main Results:

• The model learned concise latent embeddings that accurately reconstruct raw microbial growth data.
• Achieved few-shot learning for antibiotic classification and concentration prediction.
• Demonstrated accurate forecasting of microbial communities and inference of total abundance from relative abundance data.

Conclusions:

• The foundation model provides a general framework for analyzing and predicting microbial community dynamics.
• Transferable representations extracted from heterogeneous datasets improve analysis with limited measurements.
• Enables robust predictions across diverse microbial systems and applications.