iiinova — where your digital self keeps evolving

Yes that’s exactly what I meant, I think it’s unique ;)

Entrance of this helix? Or should we meet at your poster?

100

Sounds good, how about tomorrows coffee break at 10:00?

Gregor Sturm @grst.bsky.social · 13/08/2025

Would love to hear and talk about muspan and its integration into existing single cell data formats. When are you presenting your poster?

220

Reposted by Lukas Hatscher

Our benchmark + guidelines for atlas-level differential gene expression of single cells is online: academic.oup.com/bib/article/... Bottom line: Use pseudobulk + DESeq2 in simple and pseudobulk + DREAM in more complex settings. Collab w/ @leonhafner.bsky.social @itisalist.bsky.social

1156

Lukas Hatscher @loggas.bsky.social · 12/10/2025

Well this also depends on how we define the term hallucinations and the output an LLM is generating. I think it’s more terminology problem here. Which does not change the fact that a lot of people do not understand what an LLM is doing and how to proper interpret the outputs.

010

Lukas Hatscher @loggas.bsky.social · 18/09/2025

2 decades of self-injecting venom and hundreds of snake bites

000

And last but not least thanks a lot to @denisschapiro.bsky.social who established the collaboration and mentored me.

010

I am happy to have worked with my collaborators from NTNU (Ingrid and Therese) on this amazing project, as well as @chiaraschiller.bsky.social who developed COZi (www.biorxiv.org/content/10.1...). shout out to CellCharter developer @marcovarrone.bsky.social for his amazing method(and collaboration)

The spatial analysis highlight is that we uncover this signal using 2 different spatial resolutions (cell and neighborhood level) and 2 independent methods (COZI and CellCharter). We hope that this will open up new research paths in Myeloma focusing on these cell interactions.

100

To our very surprise we find that increased “interaction” of PCs and a variety of immune cells, especially CD4+Tcells, is associated to increased risk of progression, which is contrary to many findings in other tumors where tumor immune interaction seem to generally be beneficial for patients.

Lastly we apply cell neighbor preference analysis with COZI (developed by @chiaraschiller.bsky.social ) and CellCharter’s neighborhood enrichment method and connect these findings to associated clinical metadata:

100

This questions the common belief that malignant PCs solely rely on glycolytic metabolism for cancer progression and niche establishment. We further show that the aggregate size of the PC_OXPHOS neighborhood negatively correlates with immune infiltration

This led to the finding of 2 different malignant PC neighborhoods: 1) PC_OXPHOS characterized by huge vascularized aggregates of PCs with increased oxidative phosphorylation and 2)PC_MYELOID, where PCs show glycolytic metabolism and are loosely scattered around including myeloid cells.

100

As our antibody panel focused on functional markers, we used a novel neighborhood algorithm CellCharter ( @marcovarrone.bsky.social ) to structure the tissue into neighborhoods driven by not only cell types but also functional state.

We show that: MM patients with bone disease (a frequent comorbidity) show an increased abundance of malignant Plasma Cells (PCs) in the vicinity of Osteoclasts and that PCs display a bone distance dependent expression of factors involved in bone degradation (IL32, HIF1A)

100

We apply IMC to biopsies from 65 MM patients, 6 SMM and 5 MGUS patients with an antibody panel focusing on immune, bone cells and metabolism. The dataset consists of roughly 1 million labeled cells including distance to the next bone surface for every image (soon on zenodo 10.5281/zenodo.17093203)