Microscopic image analysis of Pichia pastoris based on deep learning

• 1、 School of Biotechnology, Jiangnan University, WuXi 214000

Abstract：During fermentation, changes in yeast morphology are associated with changes in cell physiology and are influenced by process conditions. At present, the monitoring of fermentation process mainly relies on physical and chemical parameters, and less attention is paid to morphological changes. Pichia pastoris cells are one of the commonly used expression systems in industry. In this study, we studied the image processing of Pichia pastoris. Microscopic images were obtained after dispersing yeast suspensions with PBS buffers and dispersants. Based on bright-field microscopic images of unstained yeast, a BAM-ResNet18 model was developed using deep learning for two-stage yeast morphological classification. The average accuracy rate of three classifications is 97.83%, further subdivided into six classifications, the accuracy rate is 92.31%. By judging the morphological changes in the yeast fermentation process, relevant information on microbial morphology and physiology can be collected. Process monitoring based on changes in cell morphology will help to enhance process understanding and control of fermentation processes.

Keywords： Pichia pastoris Microscopic image recognition Deep learning