Congratulations to @michaelgeuenich.bsky.social selected as the Canadian representative at the Young Investigator Symposium at SITC 2025.

He's presenting our work on tertiary lymphoid structures in pancreatic cancer tomorrow at 8.15am at SITC

Abstract ➡️ www.immunotherapycancer.ca/2025-michael...

Great collaboration with @gingraslab.bsky.social
& congrats to lead author Vesal.

Paper: www.nature.com/articles/s41...

Thanks to @sinaihealth.bsky.social

How can we scale subcellular spatial proteomics across many experiments and conditions?

Our new paper presents a genetic algorithm and evaluation framework that optimizes bait subsets for proximity proteomics platforms like BioID enabling context-specific proteome mapping.

Fantastic hosting @alexr.bsky.social in Toronto today discussing the hype of AI in cancer biology and a realistic route to leverage it for personalized immunotherapy

Interested in how to infer tumour microenvironment ecosystems from scRNA-seq atlases?

I'll be at #AACR25 presenting in the ED06 session this Friday 3-4.30pm.

Otherwise would love to meet to chat science for anyone on here attending

Preprint: www.biorxiv.org/content/10.1...

Foundation Models -- A Panacea for Artificial Intelligence in Pathology?

Interesting paper-training on top of FMs quickly gets you close to asymptotic performance -> could be useful to quickly assess how solvable a prediction problem is before assembling full dataset

arxiv.org/abs/2502.21264

When asked about knockout vs knockdown, Google AI wonderfully bet-hedges on whether you're interested in genetics or boxing

We developed a new metric for the biological plausibility of inferred cellular phenotypes that can be widely applied across tissues and generated imaging mass cytometry of mixture-controlled cell pellets of cell lines where the underlying phenotypes are known and act as a ground truth (3/4)

STARLING explicitly models the consequences of segmentation errors, scalably inferring both denoised cell clusters and per-cell segmentation error probabilities, all without requiring prior knowledge of cell types or abundances (2/4)

Google AI describing triple sec as “cloyingly artificial” is the real AGI milestone everyone missed

The killer AI use case nobody is talking about (and it actually works!)

💡Thinking of using scRNA-seq to design antibody panels for spatial or single-cell profiling?

Happy to launch Cytomarker, an interactive web-app and new antibody screening strategy to design and validate custom panels

💻 App: www.cytomarker.ai
📜 Preprint: www.biorxiv.org/content/10.1...

How can active and self-supervised learning methods improve the efficiency of human annotation of single-cell data?

Our paper exploring this (& more!) is now out ➡️ www.nature.com/articles/s41...

It might not be AGI but it definitely understands one-upmanship

Thank goodness cybercriminals will no longer be able to write grants and apply for funding for me