Learning Biological Dynamics From Spatio-Temporal Data by Gaussian Processes.

Area of Science:

• Computational Biology
• Systems Biology
• Machine Learning

Background:

• Understanding complex biological processes often requires intricate mathematical models.
• Current methods for biological model discovery can be challenging due to the high dimensionality and complexity of biological systems.

Purpose of the Study:

• To introduce an "equation-free" method for learning biological dynamics from spatio-temporal data.
• To demonstrate the utility of Gaussian processes for modeling biological systems without explicit model derivation.

Main Methods:

• Utilizing Gaussian processes to model biological dynamics from spatio-temporal data.
• Leveraging the local nature of biological processes for efficient learning from time-sparse data.
• Employing the squared exponential covariance function to tune hyperparameters and gain insights into underlying processes.

Main Results:

• The proposed method successfully learns biological dynamics from spatio-temporal data, even when data is sparse in time.
• Tuning hyperparameters of the squared exponential covariance function revealed key insights into the underlying biological process.
• The method demonstrated effective multi-step forecasting with simplified uncertainty propagation.

Conclusions:

• Gaussian processes provide a powerful, "equation-free" approach for discovering biological dynamics.
• This method offers a robust way to analyze complex biological systems and provides interpretable insights through hyperparameter tuning.
• The approach is validated on both synthetic data and real-world E. coli colony image data, showcasing its practical applicability.