YeastSAM: A Deep Learning Model for Accurate Segmentation of Budding Yeast Cells

Area of Science:

• Microscopy and quantitative image analysis
• Cell biology and yeast genetics
• Deep learning applications in biological imaging

Background:

• Accurate cell segmentation is crucial for quantitative image analysis.
• Budding yeast segmentation is difficult due to asymmetric division, often leading to misidentification of dividing cells.
• Existing methods struggle with the unique morphology of budding yeast.

Purpose of the Study:

• To develop an accurate and accessible deep learning framework for budding yeast cell segmentation.
• To overcome the challenge of misidentifying dividing yeast cells in microscopy images.
• To enable advanced quantitative analysis of gene expression and spatial regulation in yeast.

Main Methods:

• Development of YeastSAM, a deep learning model derived from μSAM and optimized for budding yeast.
• Training and validation of the model using microscopy images of budding yeast.
• Integration of YeastSAM with single-molecule RNA and organelle imaging techniques.

Main Results:

• YeastSAM achieves over threefold higher accuracy in segmenting dividing yeast cells compared to current methods.
• The framework successfully reduces errors caused by mother-bud morphology.
• Enables precise quantitative analysis of spatial gene expression regulation.

Conclusions:

• YeastSAM provides a high-accuracy, accessible solution for budding yeast cell segmentation.
• The tool empowers researchers, even those with limited programming experience, to perform sophisticated image analysis.
• Facilitates deeper understanding of cellular processes in budding yeast through improved quantitative imaging.