Also, in light of this, it’s probably important that the data from the Porat preprint on O-linked glycoRNA shows that Core1 O-glycans are required for the glycoRNA signal (rather than direct linkage). I could absolutely buy that losing Core1 screws with endosomal system.

Do we have a cellular system that tidies up/noms away at negatively charged polysaccharides (or polysaccharide-like RNA)?

I had a hunch about a bunch of RNA binding proteins that could pick up sugars, but absolutely did not expect all those proteins that were detected. CD44, the LAMPs, ITGB1, LRP1, CALX/CALR? Points to some pretty interesting trafficking going on with RNA around endosome/lysosome.

Peter Panum was a Dane working in Germany, and the amazing brutalist building we were based in (and are now next to) is named after him.

Questionably fun fact: The study of endotoxin in modern age was kicked off by Peter Panum (studying putrid fluids). He had also done work on albumen, releasing “something” with dilute acid hydrolysis. Eichwald in 1865 did that same reaction on mucin, arguably starting the study of glycoproteins.

This looks great - I guess the N-glyco pathway is to be expected (across all kingdoms), but what about the other glyco pathways? Is there any place a glyco nerd can look at the matched InterPro domains? It would be interesting to check for evidence of it using compartments to organise the process.

Maybe get bots to talk to each other for sharing data, and you have a much more efficient system. Why burden bots with inefficient literature based encoding of data? But all this can (and probably will) happen independently of a publishing system.

The only legitimate reason to publish is because we value the feedback and discussion with other humans for our work (yes, metrics measure our activity, but should not drive it). If I had a great robot reviewer, I wouldn’t see the point of publishing, as I would just point the bot at the raw data.

All this before even addressing the fundamental philosophical issue: do we want to value algorithmic judgment of a piece of human work? If your job is running a system, this solution looks great, but for those interacting with the system it is not quite as exciting.

They will never have this contextual information. However, personal LLMs in principle could, if the person diligently prepared the info in text format. If we don’t have time to review, we sure don’t have time to parent a private LLM.

This is due to training cut-offs, and because in human review, manuscripts are compared against the reviewers accumulated experience, which is not primarily stored in a textual format that is publicly accessible for the LLM to use. This argues against using centralised LLMs for peer review.

The issue - as I see it - is that it signals that LLMs are sophisticated enough to review work. At a technical level, LLMs are impressive (I was using one yesterday to vibe-code an interactive webpage for teaching), but their context matters. I argue LLMs will miss relevant context.

We are looking to work on the next frontiers of structural biology, where we understand the shapes and arrangements of these dynamic machines on the cell surface. Can we decode the organisational language that coordinates all these machines? (We think, yes it is possible!).

It’s a dumb idea, but what is worse, is that the Nature editor seemed to think that it had enough space to publish this fluff piece ostensibly about a random browser extension with 891 users in total today. How is this in any way interesting or relevant, apart from farming outrage clicks?

100% - it would be great with momentum in this space, and I very much appreciate how Rust is at just the right abstraction level to be useful as the backend for a lot of stuff.

Great that symmetry was an initial consideration in this, but what a shame that this fell by the wayside - GlcN/GlcA are especially bad because they are basically the same symbol. I guess once angled linkages came into the picture, it was game over for rotational invariance. Tradeoffs I guess!