Generative deep learning model assisted multi-objective optimization for wastewater nitrogen to protein conversion by...

Area of Science:

• Environmental Microbiology
• Biotechnology
• Artificial Intelligence

Background:

• Photosynthetic bacteria (PSB) are explored for wastewater nitrogen treatment and valorization.
• Balancing nitrogen removal and resource recovery in PSB remains a challenge.

Purpose of the Study:

• To optimize nitrogen removal, protein concentration, and nitrogen-to-protein conversion in PSB using generative deep learning.
• To identify key factors influencing these processes and achieve multi-objective optimization.

Main Methods:

• Generative deep learning (Variational Auto-Encoders) for data augmentation.
• Elastic Neural Network (ENN) for model fitting.
• SHapley Additive exPlanations (SHAP) for factor identification.
• Multi-objective optimization to identify Pareto front solutions.
• Validation experiments using Near Infrared (NIR) light.

Main Results:

• Generative models enhanced datasets, improving ENN model fitting.
• Key factors identified: carbon/nitrogen sources, light type, NLR, COD, and HRT.
• Multi-objective optimization yielded eight Pareto front solutions.
• Near Infrared (NIR) light significantly improved nitrogen-to-protein conversion (up to 44%) and related gene expression.

Conclusions:

• Generative AI frameworks offer innovative solutions for multi-objective optimization in wastewater nitrogen valorization, especially with limited data.
• NIR light is a promising factor for enhancing nitrogen-to-protein conversion efficiency in PSB systems.