On sale now: the Dirt Cheap J-Link Adapter
with 2x5 1.27mm and 2x10 2.54mm headers!
plugs directly into every jlink, even the mini!
supports 3v3 mods!
www.dirtcheaphackingtools.com/product/dirt...
#segger #jlink #hacking #jtag #smallbusiness #shop
All in one SEGGER J-Link adapter fits both J-Link and Mini on one convenient board www.dirtcheaphackingtools.com/product/dirt...
#jtag #swd #jlink #segger #breakout #hacking #hardware #electronics #smallbusiness
9/ #Segger grew from conferences, hackathons, & community efforts. Huge thanks to the co-lead Andrew Moorman, co-authors and colleagues, and supervisors @moritzgerstung.bsky.social and @danapeer.bsky.social . More contributions of any kind are always welcomed! & Big thanks to @scverse.bsky.social
8/ Please check our preprint (biorxiv.org/content/10.1...) and documentation (elihei2.github.io/segger_dev/), test #segger (github.com/EliHei2/segg...), and join in - PRs & issues invited!
7/ "Speed and usability first"! #Segger is built on @scverse.bsky.social, @pytorch.org, @PyG, @lightningai.bsky.social, @geopandas & @RAPIDSai to zip through big ST datasets smoothly.
Segmentation of Xenium lung cancer tissue was assessed using ground-truth staining. Cellpose served as a reference. 10x over-expanded boundaries (~180 µm²) with ~10% contamination, while Baysor over-segmented, doubling cell counts. Segger balanced precision and recall (~140 µm²), minimizing contamination (~5%) while preserving epithelial structure.
6/ On @10xgenomics.bsky.social Xenium tests (with/without membrane staining), #segger outperforms other approaches - higher sensitivity & specificity, fewer edge errors, cleaner profiles, more reliable downstream results, in a split of compute time.
segger optionally regroups unassigned transcripts based on their gene identity and spatial proximity into fragments
Some fragments and cells are shown for 2 FOVs on a Xenium breast cancer dataset.
5/ Cell-assigned transcripts form #segger 's “cells”. Optionally, the rest (up to 50%) could be grouped as “fragments”. First time to split these two entities semantically—key for handling cut-up, overlapping, or unstained cells!
#Segger embeds cells & transcripts into a shared space and assigns the cell with the highest score for each transcript within a receptive field.
4/ #segger embeds cells & transcripts in a shared latent space - closer means likelier transcript-to-cell matches. It assigns each transcript to the cell with the highest score within a radius, in the physical space.
segger's network architecture.
Schematic of #segger’s workflow: cell polygons and transcript points build a heterogenous graph. A graph attention network then propagates messages on this graph to compute transcript-to-cell association scores.
3/ #segger ’s trick? Graph Attention Networks (GATs) + GPU acceleration on heterogeneous graphs. It represents transcripts and cells as nodes of a heterogeneous graph, and computes on gene identities, transcript locations (in 3D), nucleus/membrane stains, and their morphology.
Accurate cell segmentation in ST data faces challenges like overlapping cells, transcript diffusion, background noise, and missing nuclei. Conservative assignments may over-segment cells, while under-segmentation leads to false positives that contaminate profiles. Effective segmentation must balance splitting cells and merging multiple cells.
2/ If you’ve worked with Spatial Transcriptomics (ST) data, you’ve indeed had some troubles with segmentation accuracy. Segmentation problems mess up downstream analysis & obscure cell-cell interactions - which are ST’s biggest promises. #segger
Segger logo
Message passing intuition behind the segger’s link-prediction model and the network architecture
1/ New preprint! 🍳
@elihei.bsky.social and our team at @embl.org , @dkfz.bsky.social, and @mskcancercenter.bsky.social built #segger - a fast, accurate cell segmentation tool for spatial transcriptomics that assigns transcripts to their cell origins!
doi.org/10.1101/2025...
