Mini-batch optimization enables training of ODE models on large-scale datasets.

Area of Science:

• Systems Biology
• Computational Biology
• Machine Learning

Background:

• Quantitative dynamic models are crucial for understanding cellular signal processing.
• Parameter estimation in these models is computationally intensive, especially for large-scale systems.
• Existing optimization methods struggle with growing model sizes and datasets.

Purpose of the Study:

• To adapt and benchmark mini-batch optimization for ordinary differential equation (ODE) models.
• To establish a link between dynamic modeling and machine learning techniques.
• To improve the computational efficiency and scalability of parameter estimation.

Main Methods:

• Application of mini-batch optimization, a technique from deep learning, to ODE models.
• Benchmarking against established parameter optimization approaches.
• Utilizing a large-scale cancer signaling model as a primary application example.

Main Results:

• Mini-batch optimization achieved superior results compared to established methods on a cancer signaling model.
• Computation time was reduced by over an order of magnitude.
• Demonstrated improved scaling properties for parameter optimization.

Conclusions:

• Mini-batch optimization offers a computationally efficient alternative for parameter estimation in ODE models.
• This approach enables the modeling of larger and more complex biological systems.
• The study bridges dynamic modeling and machine learning for enhanced biological insights.