Deep-learning-assisted medium optimization improves hyaluronic acid production by Streptococcus zooepidemicus

To improve the efficiency of hyaluronic acid production by Streptococcus zooepidemicus, the growth medium was optimized with a pipeline involving a deep learning (DL) algorithm. To train the DL model, the initial training dataset (OA01–18) was designed with the L18 orthogonal array, and hyaluronic acid (HA) was produced in small-scale cultures in deepwell plates. The range of HA production was 0.09–1.39 g/L under these conditions. In searching for the optimal medium composition, 54 candidate optimized media (OM01–54) were proposed by the system. According to the confirming culture experiment, the best production of HA (1.66 g/L) was achieved with OM30. During confirmation in a stirred-tank reactor, the volumetric production of HA in OA30 was larger than that in the control medium. In fed batch culture, HA accumulated to 5.13 and 9.96 g/L_{initial volume} after 10 and 30 h in culture, respectively. To avoid the suppression of HA production by the high viscosity of the medium conferred by HA, repeated batch culture with OM30 was performed by replacing 90 % of the broth volume approximately every 6 h. As a result, 21.4 g of HA was produced in 46 h, and productivity reached 0.465 g/L_{initial volume}/h.