On Wednesday, September 23, at 18:00 EEST, we are delighted to host our guest Elyas Heidari from the EMBL Heidelberg, who will present #segger,  a method for cell #segmentation from #spatial #transcriptomics.

Talk: "Segger: Fast and accurate cell segmentation of imaging-based spatial transcriptomics data"

Abstract:
Assigning transcripts to their cell of origin remains the weakest link in imaging-based spatial transcriptomics (iST), despite underpinning almost every downstream analysis. Existing methods over- and under-segment cells, misassign transcripts, and scale poorly to whole slides. We introduce Segger, a heterogeneous graph neural network that represents transcripts and cells as distinct node types and reframes segmentation as transcript-to-cell link prediction, combining transcript co-localization with nuclear or membrane staining. Three complementary training objectives align the learned transcript and cell embeddings with covariance structure estimated from the iST data themselves, so no external single-cell reference is required. Across nine datasets spanning five platform families and seven tissues, Segger gives the most balanced performance on an eight-metric suite covering sensitivity, specificity and morphological consistency, reaching lower spurious co-expression than any comparator at matched marker recall, with a single user-facing parameter that traverses the trade-off. Because Segger preserves staining-derived geometry, it also yields faithful cell–cell neighbourhood graphs, whereas molecularly coherent but geometrically implausible assignments distort them. In healthy human colon, Segger recovers neutrophils and an inflammatory epithelial niche that nucleus-only segmentation misses entirely. Segger processes full slides of 10⁷ to 10⁹ transcripts within practical runtime and memory on a single GPU, and is available as documented open-source software at https://github.com/dpeerlab/Segger.

Keywords: spatial transcriptomics, cell segmentation

Resources:

• Segger Github https://github.com/dpeerlab/segger
• Segger pre-print https://www.biorxiv.org/content/10.1101/2025.03.14.643160v1
• Elyas Heidari https://elihei2.github.io/

Format: Online, free to attend, requires registration: https://luma.com/pv4gkoaf

Organization: YMS is organized by RSBI and is linked to RoBioinfo Seminars

Chair: Madalina Giurgiu-Kraljic

Click here to subscribe for upcoming #YMS Seminars and #RSBI Events

Check all bioinformatics events on CRB website