Grey-box modeling framework for consolidated bioprocessing systems: an endpoint-guided approach

Area of Science:

• Biotechnology and Bioengineering
• Process Systems Engineering
• Computational Biology

Background:

• Consolidated bioprocessing (CBP) integrates enzyme production, biomass hydrolysis, and fermentation, but faces modeling challenges due to biological nonlinearities, feedstock variability, and sparse data.
• Accurate modeling is crucial for optimizing CBP efficiency and yield in biofuel and biochemical production.

Purpose of the Study:

• To develop and validate an endpoint-guided grey-box framework for predicting CBP outcomes.
• To reconstruct mechanistic process behavior and assess state estimation under data-scarce conditions.

Main Methods:

• Utilized a literature-derived CBP ethanol dataset (540 runs, 90 features) with out-of-fold residual screening for preprocessing.
• Employed and compared nonlinear regressors, selecting an XGBoost model (XGB_raw) for endpoint prediction.
• Coupled the endpoint surrogate with a three-phase hybrid simulator and performed synthetic state-estimation analysis using an unscented Kalman filter.

Main Results:

• The XGB_raw model, after residual screening, achieved improved prediction accuracy (RMSE [Formula: see text]).
• Feature attribution identified key predictors: hemicellulose, substrate concentration, temperature, residence time, enzyme pretreatment, pH, cellulose, and mixing rate.
• The framework successfully reconstructed plausible biological parameterizations and demonstrated accurate state estimation for sugars and products, with moderate fidelity for enzyme dynamics.

Conclusions:

• The endpoint-guided grey-box framework offers a feasible approach for CBP endpoint prediction and mechanistic interpretation.
• The study provides a basis for preliminary soft-sensing system design in CBP under sparse-data constraints.
• This methodology addresses critical gaps in modeling complex bioprocesses, enhancing potential for optimization and control.