Bayesian uncertainty quantification for anaerobic digestion models.

Area of Science:

• Computational biology
• Environmental engineering
• Statistical modeling

Background:

• Uncertainty quantification is crucial for reliable computational models in biology.
• Anaerobic digestion models are widely used but require robust uncertainty assessment.
• Existing methods for uncertainty quantification can be computationally expensive or overly confident.

Purpose of the Study:

• Introduce a novel generalized Bayesian procedure (VarBUQ) for uncertainty quantification.
• Evaluate VarBUQ's performance against established methods using synthetic data.
• Provide a computationally efficient Bayesian approach for biological models.

Main Methods:

• Developed a generalized Bayesian procedure named VarBUQ.
• Benchmarked VarBUQ against Fisher's information, bootstrapping, and Beale's criteria.
• Utilized synthetic data for comparative analysis of uncertainty quantification methods.

Main Results:

• VarBUQ demonstrated a favorable balance between model fitting and confidence estimation.
• Traditional methods (Fisher's information, bootstrapping, Beale's criteria) were found to be excessively confident.
• The performance of VarBUQ was enhanced by the inductive bias of a carefully constructed prior distribution.

Conclusions:

• VarBUQ offers a computationally efficient and reliable method for uncertainty quantification in anaerobic digestion models.
• The study advocates for increased attention to uncertainty in biological modeling.
• A Python package 'aduq' is released to support the implementation of VarBUQ.