Efficient fermentation process for rhamnolipid production based on machine vision and Bayesian optimization

Abstract:

Rhamnolipids, a class of biosurfactants, are widely applied in industries such as petrochemicals, environmental agriculture, daily chemicals and pharmaceuticals. However, their industrial application is severely constrained by the low rhamnolipid yield of natural strains and high overall production costs. To obtain microbial strains with high rhamnolipid-producing capacity and to enhance their fermentation yields, this study developed a screening method for rhamnolipid-producing microorganisms assisted by the oil spreading technique based on machine vision (MV). Using this approach, a Pseudomonas aeruginosa strain PA022 with favorable rhamnolipid-producing capacity was isolated from an environmental soil sample. Based on single-factor preliminary experiments, Bayesian optimization (BO) coupled with Latin hypercube sampling (LHS) was applied to perform global optimization of key fermentation parameters, thereby determining the medium composition and culture conditions for PA022 as follows: rapeseed oil 30 g/L, NaNO₃ 12 g/L, peptone 3 g/L, phosphate at a ratio of 3:1 with a total concentration of 1.2 g/L, initial pH 6.8, temperature 35 ℃, and inoculum size 2.5%. Under the optimal fermentation conditions, strain PA022 achieved a rhamnolipid titer of 20.01 g/L in shake-flask fermentation, and the yield reached 30.52 g/L at the 5 L fermenter scale. This study demonstrates that the small-sample global optimization strategy combining MV and BO can enhance the efficiency of strain screening and fermentation process development, thereby providing a reference for the optimization of rhamnolipid and analogous fermentation processes.