Predicting the performance of anaerobic digestion using machine learning algorithms and genomic data.

Area of Science:

• Environmental microbiology
• Biotechnology
• Data science

Background:

• Anaerobic digestion (AD) modeling is vital for optimizing digester performance but is challenging due to complex system interactions.
• Data mining and microbial gene sequencing offer promising solutions for understanding and improving AD processes.

Purpose of the Study:

• To evaluate the effectiveness of six machine learning (ML) algorithms in predicting methane yield using genomic and operational data.
• To identify key microbial phyla influencing methane production in anaerobic digestion.

Main Methods:

• Utilized genomic data (bacterial phylum level) and operational parameters from 8 research groups.
• Applied and compared six ML algorithms for classification and regression tasks.
• Performed feature importance analysis using Random Forest (RF) to identify significant microbial taxa.

Main Results:

• Random Forest (RF) classification achieved 0.82 accuracy combining operational and genomic data.
• Neural network regression yielded a low root mean square error of 0.04 using genomic data alone.
• Identified Chloroflexi, Actinobacteria, Proteobacteria, Fibrobacteres, and Spirochaeta as key phyla impacting methane yield.

Conclusions:

• ML techniques, particularly RF and neural networks, show significant promise for predicting and controlling anaerobic digestion performance.
• Genomic data, even at the phylum level, provides valuable insights for AD process optimization.
• Identified microbial communities can guide proactive management and early warning systems for digester operations.