Morphological classification of Schizochytrium and mutagenic selection of high-oil-producing strains based on deep learning.

As a natural producer of omega-3 fatty acids, Schizochytrium demonstrates exceptional cell density and docosahexaenoic acid (DHA) production efficiency, establishing its status as a microbial platform of industrial significance. However, the absence of real-time fermentation monitoring systems constrains production optimization and hinders the sustainable commercialization of omega-3 products. Morphometric analysis offers potential for tracking microbial growth and metabolite accumulation, yet the complex relationships between Schizochytrium morphology, biomass dynamics, and lipid biosynthesis remain unresolved. Conventional morphological characterization relies on labor-intensive microscopic observation by trained personnel, necessitating automated image analysis solutions. We developed a novel 13-class morphotype classification system integrating cellular division characteristics and lipid droplet parameters, coupled with a purpose-built object detection architecture. The enhanced MLC-YOLO framework achieved 84.2 % mean average precision (mAP), representing a 2.2 % improvement over the standard YOLOv8s implementation. Fermentation monitoring identified strong positive correlations (p < 0.001) between Lipid-saturated unicells (G11) and lipid yield, whereas Small lipid droplet tripartite cells (G4) and Small lipid droplet quadripartite cells (G6) inversely correlated with productivity. Application of the G11/(G4 +G6) selection index facilitated the isolation of mutant strain S62, which had a lipid content of 50.46 %, an increase of 1.96 % over the parental strain. This study addresses fundamental knowledge gaps in Schizochytrium morphology, establishes deep learning-enabled cellular phenotyping as a viable strain selection strategy, and propels the development of smart biomanufacturing systems for industrial omega-3 production.