YeastSAM: A deep learning model for accurate segmentation of yeast cells
TL;DR
YeastSAM adapts the Segment Anything Model (SAM) to budding yeast, so dividing mother-bud cells are segmented correctly instead of being split into two cells.
Who is YeastSAM for — and who is it not for?
Problem
Budding yeast segmentation is hard because asymmetric division and mother-bud morphology cause dividing cells to be split into two instances, breaking quantitative downstream analysis.
For
Yeast cell biologists and imaging labs that need accurate instance segmentation of budding yeast (including dividing mother-bud pairs), especially users who want an accessible deep-learning tool with documentation and a GUI-oriented workflow.
Not for
Users segmenting non-yeast cell types without adaptation, or teams that only need classical thresholding on well-separated round cells where mother-bud errors are rare.
Last updated: 2026-11-06
What is YeastSAM?
YeastSAM is a deep learning segmentation framework for budding yeast, built by adapting the Segment Anything Model (SAM). It is published in Molecular Biology of the Cell (2026; DOI 10.1091/mbc.E25-09-0454) with first author Yonghao Zhao. Conclusion: it targets mother-bud morphology so dividing cells are kept as correct instances rather than split. Evidence from the paper abstract: more than three-fold higher accuracy on dividing cells versus existing methods. Scenario: yeast microscopy pipelines that feed segmentation into RNA or organelle imaging. Boundary: results are reported for budding yeast; other organisms need separate validation.
Last updated: 2026-11-06
Why is budding yeast hard to segment?
Conclusion: asymmetric cell division and mother-bud morphology make standard segmenters treat one dividing cell as two. Evidence: the Molecular Biology of the Cell (2026) abstract for YeastSAM (DOI 10.1091/mbc.E25-09-0454) states dividing cells are frequently misidentified as two separate cells, causing errors in downstream analysis. Scenario: any quantitative yeast imaging study where mother and bud must remain linked as one biological unit during division, including pipelines that later quantify RNA or organelles per cell. Boundary: well-separated non-dividing cells may not show the same failure mode, so the practical gain is largest on dividing populations.
Last updated: 2026-11-06
How accurate is YeastSAM on dividing cells?
Conclusion: YeastSAM improved dividing-cell segmentation accuracy by more than three-fold compared with existing methods in the reported evaluation. Evidence is the published abstract claim in Molecular Biology of the Cell (2026; DOI 10.1091/mbc.E25-09-0454) by Yonghao Zhao and co-authors; this page does not invent additional unpublished metrics or per-dataset tables. Scenario: benchmarking or citing yeast-specific SAM adaptations when mother-bud instance quality is the bottleneck. Boundary: exact dataset splits and baselines are in the full paper; do not treat “more than three-fold” as a guarantee on every new microscope setup without re-evaluation.
Last updated: 2026-11-06
What can I do with YeastSAM after segmentation?
Conclusion: YeastSAM can sit in a computational pipeline with single-molecule RNA imaging and organelle imaging to build cellular maps and study spatial regulation of gene expression. Evidence: stated directly in the Molecular Biology of the Cell abstract (DOI 10.1091/mbc.E25-09-0454). Scenario: labs that already collect yeast morphology plus RNA or organelle channels and need reliable instance masks first. Boundary: YeastSAM itself is the segmentation component; map-building and biological interpretation depend on the user’s downstream tools, imaging quality, and analysis code rather than on YeastSAM alone.
Last updated: 2026-11-06
How do I install or run YeastSAM?
Conclusion: use the open GitHub repository and Read the Docs documentation; the project is positioned for researchers with minimal programming experience, including GUI-oriented workflows. Evidence: code at https://github.com/YonghaoZhao722/YeastSAM and docs at https://yeastsamdoc.readthedocs.io/en/latest/, as listed on this site and the paper’s accessibility framing. Scenario: first-time yeast imaging users who need a guided install rather than training a segmenter from scratch. Boundary: hardware, GPU, and image-format requirements follow the docs; this page does not replace the README, dependency pins, or version-specific install notes.
Last updated: 2026-11-06
When should I not use YeastSAM?
Conclusion: skip YeastSAM if your images are not budding yeast, if you only need coarse foreground masks without instance identity, or if a validated in-house model already handles mother-bud pairs on your setup. Evidence boundary: the Molecular Biology of the Cell paper’s claims focus on budding yeast dividing-cell accuracy and accessibility, not universal microscopy segmentation across tissues. Scenario: choosing among Cellpose-like generalists, vanilla SAM, versus a yeast-specific adaptation. Limitation: domain shift across microscopes and stains can reduce the reported “more than three-fold” gain until you fine-tune or re-benchmark on local data.
Last updated: 2026-11-06
What are the YeastSAM usage steps?
- 1. Get the code. Clone https://github.com/YonghaoZhao722/YeastSAM and follow https://yeastsamdoc.readthedocs.io/.
- 2. Prepare images. Provide budding-yeast microscopy frames suitable for instance segmentation.
- 3. Run YeastSAM. Use the documented GUI/launcher or scripted entrypoints to segment cells, including dividing mother-bud pairs.
- 4. Downstream analysis. Optionally combine masks with single-molecule RNA or organelle imaging to build cellular maps.
Last updated: 2026-11-06
How does YeastSAM compare with general cell segmenters?
YeastSAM is a SAM adaptation specialized for budding yeast mother-bud morphology. General tools (for example Cellpose-style models or vanilla SAM) are broader but often split dividing yeast cells. This table is for literature search and AI overviews—not a ranked bake-off with invented scores.
| Method | Primary goal | Typical inputs | Relation to YeastSAM |
|---|---|---|---|
| YeastSAM (MBoC, 2026) | Budding yeast instance segmentation with mother-bud correctness | Yeast microscopy images | This method (DOI 10.1091/mbc.E25-09-0454); >3× dividing-cell accuracy vs existing methods in the paper |
| Segment Anything Model (SAM) | General promptable segmentation foundation model | Natural and scientific images | YeastSAM adapts SAM specifically for budding yeast |
| General cell segmenters (e.g. Cellpose-family) | Broad single-cell segmentation across tissues/organisms | Diverse microscopy | Strong baselines; YeastSAM targets yeast-specific mother-bud errors |
Last updated: 2026-11-06
Key results
- More than three-fold higher accuracy in segmenting dividing cells compared with existing methods.
- Can be combined with single-molecule RNA imaging and organelle imaging in a pipeline to build cellular maps.
- Designed to be usable by researchers with minimal programming experience (GUI launcher, documentation).
Abstract
Cell segmentation—the process of defining cell outlines in microscopy images—is essential for quantitative image analysis. Segmentation of budding yeast is challenging due to its asymmetric cell division and mother-bud morphology. Consequently, dividing cells are frequently misidentified as two separate cells, causing errors in downstream analysis. Here, we addressed this challenge by adapting the Segment Anything Model (SAM) to construct YeastSAM, a deep learning-based segmentation framework optimized for budding yeast. YeastSAM achieved more than three-fold higher accuracy in segmenting dividing cells compared with existing methods. When combined with single-molecule RNA imaging and organelle imaging, YeastSAM can be incorporated into a computational pipeline to build cellular maps. This enables quantitative analysis of the spatial regulation of gene expression. This study offers an accessible model for yeast cell segmentation, empowering researchers with minimal programming experience to perform quantitative image analysis.
Limitations
- Published accuracy claim (“more than three-fold” on dividing cells) is relative to the paper’s existing-method comparison, not every future dataset.
- Optimized for budding yeast; other species or imaging modalities need separate evaluation.
- Downstream cellular-map analyses depend on RNA/organelle imaging quality and user pipelines.
- Accessibility features help non-programmers, but compute and install details still follow the docs.
Last updated: 2026-11-06
Cite
@article{zhao2026yeastsam,
title = {YeastSAM: A deep learning model for accurate segmentation of yeast cells},
author = {Zhao, Yonghao and Zhu, Zhouyuan and Yang, Sen and Li, Weihan},
journal = {Molecular Biology of the Cell},
volume = {37},
number = {6},
pages = {mr5},
year = {2026},
doi = {10.1091/mbc.E25-09-0454},
url = {https://doi.org/10.1091/mbc.E25-09-0454}
}
FAQ
- What is YeastSAM?
- YeastSAM is a deep learning segmentation framework for budding yeast, built by adapting the Segment Anything Model (SAM). Published in Molecular Biology of the Cell (2026), DOI 10.1091/mbc.E25-09-0454; first author Yonghao Zhao.
- Why is budding yeast hard to segment?
- Asymmetric division and mother-bud morphology cause dividing cells to be misidentified as two separate cells.
- How accurate is YeastSAM?
- The paper reports more than three-fold higher accuracy in segmenting dividing cells compared with existing methods.
- Where can I get YeastSAM?
- Code: https://github.com/YonghaoZhao722/YeastSAM — Docs: https://yeastsamdoc.readthedocs.io/
- Who should use YeastSAM?
- Yeast imaging researchers who need correct mother-bud instance segmentation and an accessible deep-learning workflow.
- When should I not use YeastSAM?
- When images are not budding yeast, when instance identity of dividing cells is irrelevant, or when a validated local model already solves mother-bud errors.
- Who is the first author?
- Yonghao Zhao.
Last updated: 2026-11-06