Synthetic data-augmented machine learning approaches for tailor-made microbial conversion of methane to phytoene.

Area of Science:

• Microbial Biotechnology
• Synthetic Biology
• Metabolic Engineering

Background:

• Metabolic engineering for valuable compound biosynthesis is often limited by trial-and-error methods.
• Optimization of non-model organisms like Methylocystis sp. MJC1 presents unique challenges due to data scarcity.
• Phytoene biosynthesis from methane offers a sustainable bioproduction route.

Purpose of the Study:

• To develop a machine learning (ML)-assisted predictive framework for optimizing phytoene biosynthesis in Methylocystis sp. MJC1.
• To enhance ML model performance using synthetic data generation for non-model organisms.
• To systematically engineer the methylerythritol 4-phosphate (MEP) and carotenoid pathways for improved phytoene yield.

Main Methods:

• Developed an ML-assisted predictive framework incorporating synthetic data generation (CTGAN).
• Targeted key genes (dxs, crtE, crtB) in MEP and carotenoid pathways, modulating promoter strengths.
• Employed deep neural networks (DNN) and support vector machines (SVM) for predicting optimal promoter-gene combinations.
• Utilized conditional tabular generative adversarial networks (CTGAN) to generate synthetic data, overcoming limitations with non-model organisms.

Main Results:

• The ML-guided engineered strain showed a 2.2-fold improvement in phytoene production.
• Phytoene content increased 1.5-fold in the engineered strain compared to the base strain.
• Validated the effectiveness of the ML framework and synthetic data generation for pathway optimization.

Conclusions:

• Integrated ML-driven predictive frameworks with metabolic engineering enable rapid and precise optimization of microbial bioconversion.
• The developed approach successfully enhanced phytoene biosynthesis from methane using a non-model methanotroph.
• This study demonstrates a powerful strategy for sustainable bioproduction using engineered microbes.