Optimization of rifamycin B fermentation in shake flasks via a machine-learning-based approach

Area of Science:

• Biotechnology
• Microbial Fermentation
• Machine Learning Applications

Background:

• Rifamycin B is a vital polyketide antibiotic for treating tuberculosis and leprosy.
• Optimizing fermentation media is crucial for enhancing antibiotic production efficiency.
• Amycolatopsis mediterranei S699 is a key strain for Rifamycin B biosynthesis.

Purpose of the Study:

• To optimize the fermentation medium for enhanced Rifamycin B production using machine learning.
• To explore the efficacy of Genetic Algorithm (GA), Neighborhood Analysis (NA), and Decision Tree (DT) for medium optimization.
• To identify novel medium compositions leading to significant improvements in Rifamycin B yield.

Main Methods:

• Utilized a barbital-insensitive mutant strain (Amycolatopsis mediterranei S699).
• Applied machine learning algorithms: Genetic Algorithm (GA) for global search, Neighborhood Analysis (NA) for local search and prediction.
• Conducted 178 shake-flask experiments across five generations to explore a large combinatorial search space of nine medium components.

Main Results:

• Achieved significant optimization within five generations using GA and NA.
• Identified 11 distinct medium combinations yielding over 600% improvement in Rifamycin B productivity.
• GA demonstrated superior performance in optimizing the fermentation medium compared to NA.
• Decision Tree technique provided qualitative insights into media-media interactions affecting productivity.

Conclusions:

• Machine learning, particularly GA, is highly effective for optimizing complex fermentation media.
• The identified optimal media compositions offer a substantial increase in Rifamycin B production.
• This study provides a foundation for further enhancing antibiotic manufacturing processes through data-driven approaches.