Efficient data collection for establishing practical identifiability via active learning.

Area of Science:

• Bioengineering
• Systems Biology
• Mathematical Modeling

Background:

• Practical identifiability analysis (PIA) is vital for ensuring reliable parameter estimates in bioengineering models.
• Optimizing experimental design is crucial for minimizing costs and resource consumption in model development.

Purpose of the Study:

• To introduce E-ALPIPE, a sequential active learning algorithm for efficient experimental design.
• To enhance practical identifiability in bioengineering models by recommending optimal data collection points.

Main Methods:

• Developed E-ALPIPE, a sequential active learning algorithm.
• Evaluated E-ALPIPE against benchmark and random sampling methods using three synthetic experiments.
• Assessed algorithm performance based on data requirements, confidence intervals, and parameter estimate accuracy.

Main Results:

• E-ALPIPE substantially reduces the number of observations needed to achieve practical identifiability.
• The algorithm yields comparable or narrower confidence intervals compared to existing methods.
• E-ALPIPE provides more accurate point estimates of system dynamics.

Conclusions:

• E-ALPIPE offers an efficient approach to experimental design for achieving practical identifiability.
• The algorithm optimizes data collection, leading to cost and resource savings in bioengineering.
• E-ALPIPE improves the reliability and accuracy of parameter estimation in complex models.