MCseg: AI agent-guided workflow search for no-code cell segmentation and transcript attribution in spatial transcriptomics
- 1Department of Biomedical Engineering, College of Medicine and College of Engineering, National Taiwan University, Taipei 106, Taiwan
Abstract
Cell-level analysis of high-resolution spatial transcriptomics depends on accurate segmentation and transcript assignment, yet current workflows often trade transcript capture for boundary purity and can require substantial image-analysis expertise. We developed MCseg, a downloadable no-code platform whose fixed segmentation engine was derived by an AI-agent-guided search in which an AI agent iteratively proposed and evaluated combinations of image-processing and segmentation operations against Xenium-derived cell boundaries. In a lung adenocarcinoma development set, fixed-parameter MCseg increased mean panoptic quality from 0.432 to 0.472 relative to an Optuna-tuned two-diameter Cellpose baseline, while a reference-guided calibration analysis reached 0.554. In an independent expert-annotated colorectal cancer region, MCseg showed higher lineage recall and micro-F1 than the StarDist-based ENACT workflow among cells covered by both methods. Across 15 colorectal cancer regions, MCseg increased neighborhood expression discordance and reduced lineage-exclusive co-expression relative to Space Ranger at similar UMI density. The fixed workflow also transferred to fresh-frozen breast cancer without tissue-specific architecture search, illustrating an agent-guided route to reproducible, locally deployable cell-level spatial transcriptomic analysis.
Related
- Unified cell segmentation for spatial transcriptomics via transcript reassignment
- Imaging-based spatial isoform transcriptomics resolves transcript diversity at cellular and subcellular resolution
- Spatialsnake: a flexible workflow for spatial transcriptomics analysis across platforms
- Poly Pipeline: A Polyvalent Spatial Transcriptomics Workflow Validated Across Polyploid and Diploid Organisms
- Agentic AI enables cell segmentation-free cell-type inference and cell-type-guided cell segmentation in spatial transcriptomics
BibTeX
@misc{chan2026mcseg,
title = {{MCseg: AI agent-guided workflow search for no-code cell segmentation and transcript attribution in spatial transcriptomics}},
author = {C.-R. Chan and N.-W. Chang and C.-Y. Wang and H.-Y. Tan and S.-J. Lin},
howpublished = {bioRxiv},
year = {2026},
month = sep,
doi = {10.64898/2026.09.20.752837},
url = {https://www.biorxiv.org/content/10.64898/2026.09.20.752837v1},
note = {CC BY-NC-ND},
}
RIS
TY - UNPB
TI - MCseg: AI agent-guided workflow search for no-code cell segmentation and transcript attribution in spatial transcriptomics
AU - C.-R. Chan
AU - N.-W. Chang
AU - C.-Y. Wang
AU - H.-Y. Tan
AU - S.-J. Lin
PY - 2026
DA - 2026/09/20
PB - bioRxiv
DO - 10.64898/2026.09.20.752837
UR - https://www.biorxiv.org/content/10.64898/2026.09.20.752837v1
AB - Cell-level analysis of high-resolution spatial transcriptomics depends on accurate segmentation and transcript assignment, yet current workflows often trade transcript capture for boundary purity and can require substantial image-analysis expertise. We developed MCseg, a downloadable no-code platform whose fixed segmentation engine was derived by an AI-agent-guided search in which an AI agent iteratively proposed and evaluated combinations of image-processing and segmentation operations against Xenium-derived cell boundaries. In a lung adenocarcinoma development set, fixed-parameter MCseg increased mean panoptic quality from 0.432 to 0.472 relative to an Optuna-tuned two-diameter Cellpose baseline, while a reference-guided calibration analysis reached 0.554. In an independent expert-annotated colorectal cancer region, MCseg showed higher lineage recall and micro-F1 than the StarDist-based ENACT workflow among cells covered by both methods. Across 15 colorectal cancer regions, MCseg increased neighborhood expression discordance and reduced lineage-exclusive co-expression relative to Space Ranger at similar UMI density. The fixed workflow also transferred to fresh-frozen breast cancer without tissue-specific architecture search, illustrating an agent-guided route to reproducible, locally deployable cell-level spatial transcriptomic analysis.
ER -