Modeling and optimization of bioproduct formation with purple phototrophic bacteria using machine learning

Area of Science:

• Biotechnology and microbial engineering
• Bioprocess engineering
• Machine learning applications in bioprocess optimization

Background:

• Purple phototrophic bacteria offer a sustainable route for converting municipal and industrial wastewater, along with organic waste, into valuable bioproducts.
• Efficiently producing diverse bioproducts requires precise control over numerous operational parameters.

Purpose of the Study:

• To compare the performance of Random Forest, XGBoost, and CatBoost machine learning models in predicting and optimizing the synthesis of key bioproducts.
• To identify critical factors influencing the production of polyhydroxybutyrate, polyhydroxyvalerate, 5-aminolevulinic acid, coenzyme Q10, carotenoids, bacteriochlorophylls, and biomass.
• To determine optimal operating conditions for maximizing the yield of individual bioproducts.

Main Methods:

• Training and tuning machine learning models (Random Forest, XGBoost, CatBoost) using Bayesian optimization on a dataset from previous studies.
• Evaluating model performance using R², Pearson correlation, RMSE, and MAPE.
• Utilizing feature importance analysis to identify key input variables for bioproduct synthesis.
• Applying Particle Swarm Optimization to find optimal conditions for each target compound.

Main Results:

• CatBoost demonstrated superior predictive accuracy and lower error rates compared to Random Forest and XGBoost.
• Key factors influencing bioproduct formation include reaction time, mineral medium concentration, and volume exchange percentage.
• Predicted maximum yields were achieved for polyhydroxybutyrate (569 mg/L), polyhydroxyvalerate (45 mg/L), 5-aminolevulinic acid (79 µmol/L), coenzyme Q10 (13 mg/g dw), carotenoids (7 mg/g dw), bacteriochlorophylls (17 mg/g dw), and biomass (2040 mg/L).

Conclusions:

• Machine learning, especially CatBoost, is effective for predicting and optimizing bioproduct synthesis by purple phototrophic bacteria.
• Optimal bioproduct yields are achieved under specific conditions, often involving sequencing batch reactor operation and discontinuous illumination.
• Distinct operational settings are required for different bioproducts, suggesting potential for targeted production strategies and compound clustering.