iiinova — where your digital self keeps evolving

Hiren Joshi @glyco.me · 19/09/2026

This type of guy has a profoundly nostalgia-informed worldview, that completely clouds his ability to understand the current state of the world. I imagine the “This is fine” meme, only with the dog swaddled in a comfort blanket.

Hiren Joshi @glyco.me · 16/09/2026

I got my robot buddy to port over the amazing Fiziko MetaPost library to SVG. It has been on my wish list for ages.

000

Hiren Joshi @glyco.me · 16/09/2026

I was tempted to post the black and white version of this online, but we have enough conflict in the world. Fans of 6000 word essays on engraved elevator buttons in New York will appreciate the font.

210

Hiren Joshi @glyco.me · 12/09/2026

Oh right, and we have to do “innovation”. Which strikes me as pure hypocrisy as the ideas from the top show a distinct lack of organisational innovation. Same tired, failed ideas tried again and again.

Hiren Joshi @glyco.me · 12/09/2026

The only real moat we have as teachers is that the physical presence and attention we can give enables learning. At our Uni, the insanity continues as our time is eaten up by more admin, and teaching is severely under-resourced. The board would do well to remember why universities exist.

120

Yeah, it definitely reads as a pitch. Are you an invertebrate legislator lacking the anatomy to stand up to Fauci witchhunts? Protect yourself with Anthropic GrantSafe. Eliminate those pesky paper trails and automate your political influence on where you deliver your donor bonuses (grants)!

000

Worst crime in my view: the website breaks the scroll-bar so you don’t know that there is actually much more to read. However, I read the “detail” now and it is not particularly damning. I suspect the biological descriptions are Claude generated.

100

Maybe design of proteins/chemicals is an issue, but that still is hard work, and for adapting viruses it still seems like old school threats (mix animals and humans in one environment) are still much more concerning.

111

At the same time, it looks like the worst crime in Anthropics eyes was to avoid the region blocking of capabilities. What’s the argument here? They want to block unsophisticated actors from developing bioweapons? Like somehow the grant applications/bioinformatics was the rate limiting step.

100

Countering misuse of AI: September 2026 / Anthropic

The 5 use-cases tell a story. Two of them are grant applications, I find that hard to see as threatening. So then there is avian flu adaptation, the peptide atlas and toxin redesign. I can see how legitimate lines of investigation could have triggered this.

anthropic.com

Case studies from threat actors disrupted between December 2025 and August 2026 across seven areas of harm, from cyber operations to biological misuse.

121

Hiren Joshi @glyco.me · 07/09/2026

Oh the replies I could write.. but shouldn’t.

001

Hiren Joshi @glyco.me · 07/09/2026

Unfortunately not much actionable here, but I have recently begun to wonder how we can support bringing more countries (and the respective talent pools!) onto the pitch. Hard to solve funding issue from here, but training is more tractable?

Hiren Joshi @glyco.me · 29/08/2026

The scientific publishing system brings to mind the insects that are controlled by the invasive entomopathogenic fungi.

Hiren Joshi @glyco.me · 27/08/2026

The instruments do exactly what they are meant to do - its more the series of compromises needed to squeeze biological sample complexity through that tiny needle, meaning you end up with biases in your data, often really difficult to control for.

Hiren Joshi @glyco.me · 27/08/2026

Biggest challenge I face from day-to-day: what really passes as ground truth? Mass spec? Westerns? Cell lines? KO animal models? I can tell stories about how all these methods lost my unshakable trust in them. Danger zone in comp when building models of proxies/abstractions with uncorrelated noise.

Tanmay Bharat @tbharat-lab.bsky.social · 15/07/2026

Reposted by Hiren Joshi

Structural basis of biofilm formation mediated by the Pseudomonas aeruginosa fibrillar adhesin CdrA led by @olivia--smith.bsky.social Collaboration with @alexbateman1.bsky.social , Andres Floto and @geiselbiofilm.bsky.social labs

28535

Hiren Joshi @glyco.me · 18/07/2026

This one will be fun to try out!

Hiren Joshi @glyco.me · 04/07/2026

There’s a renaissance in optimised libraries for different tasks these days, so the sky is the limit with respect tohow much more intelligence can be baked into the approach. Looking forward to see it develop!

Hiren Joshi @glyco.me · 04/07/2026

I don’t know if you need to squeeze even more performance out, but I have had good experiences with Joshua Klein’s mzpeaks suite of Rust libraries. After Rust learning curve, it gives a nice way to add compiler optimised code in for a boost in speed.

Hiren Joshi @glyco.me · 04/07/2026

About bloody time someone tried doing this, that IAPI has been begging to be used in an application like this. Keep up the great work Chris (and team)! What’s the time budget you need to make glyco decisions in? Hundreds of milliseconds per precursor?

120

Hiren Joshi @glyco.me · 02/07/2026

Mark it down in the history books, 2nd July 2026 is when humanity pushed ourselves over the threshold into the AI shit-ularity.

Hiren Joshi @glyco.me · 02/07/2026

The joke, for people not following along, is that 3 pages of Claude analysis of a post is about as worthy as 100 pages of a AI generated review.

Hiren Joshi @glyco.me · 02/07/2026

Claude has learnt that apparently all I care about is glyco, which isn’t entirely unfair.

Hiren Joshi @glyco.me · 25/06/2026

The new owners LabX started in lab equipment, but perhaps more worryingly, also decided that IFLScience was worth buying. I doubt they have culture or backbone to do the right thing long term.

Hiren Joshi @glyco.me · 10/06/2026

Well, it was fun while it lasted. I’m going to have to retire this plugin as PyMOL stopped bundling the browser engine. It was so easy adding glycans to random proteins with the GUI, but alas, no more.

000

Hiren Joshi @glyco.me · 06/06/2026

I had a funny conversation yesterday when I told them about FoldDISCO and we landed on what the universal mime movement for disco is - the staying alive point.

130

Hiren Joshi @glyco.me · 06/06/2026

FoldDISCO is one of my favourite tools, I am constantly telling people they should try it out, alongside foldseek. I think you need to develop a bit of an intuition to navigate results (i.e., apply fuzzy scoring on top of calculated metrics), but once you do that it is amazing.

140

Hiren Joshi @glyco.me · 22/05/2026

There’s also the related, but impossible to systematise phylogenies of ideas/approaches across different labs - how related is the breadth of an academic lineage in a field (broad = diverse) to the probabilities of innovative ideas forming where branches cross?

Hiren Joshi @glyco.me · 19/05/2026

Who would have thought you could use so many tokens getting Claude to do the hard work to ingest GEO datasets. The scRNA-seq world is wild.

000

Hiren Joshi @glyco.me · 17/05/2026

LLMs have always been excellent when used as glue to stick different things together. Sometimes they accidentally bake some data in, but most of the time, you need to be explicit about what it glues to avoid wasting tokens. Treating it as a god machine is a waste of money.

Hiren Joshi @glyco.me · 14/05/2026

Big fan of unit tests for these tools to cover diverse/weird inputs! I am an even bigger fan of specifications that you can just follow (pop question: where is glycoworkbench sequence format specified? A: in the java code :| )!

I love this concept so much. Haven’t checked out the architecture, but I bet its a bunch of wasm running this. I notice that Claude is on the contributors on github. The tool has really been transformational in making wasm available to me personally, as it generated standard tooling.

030

Actually still don’t get it.. if I want to author quarto documents, how can I spin up a gui for editing them (with inline images a la a rstudio/jupyter docker container) on a remote server? Do the editors all live locally? Will I be forced to use positron/vscode?

000

Ipynb is a terrible format for git, and you end up missing ggplot (still great, despite all its flaws). It might be time to give quarto another go - I couldn’t figure out how to run it within a devcontainer (how we isolate our environments per project), but to be fair jupyter wasn’t dead simple too.

100

Hiren Joshi @glyco.me · 07/05/2026

I have recently ventured deep into python notebook land these days, and I would love to know how people handle moderately complex notebooks. For simple stuff code+output is easy. But anything larger and it becomes quickly unreadable. Jupyter code hiding/folding is completely inscrutable…

100

It was a pleasure to host! It was a superb double bill on N-glycosylation and all of its wonderful complexities!

Hiren Joshi @glyco.me · 03/05/2026

Yeah I have had a read of the paper since, and it is deep into speculation land there with that suggestion. I am putting it into the weird, if true bucket, and waiting for some better evidence to turn up.

130

Hiren Joshi @glyco.me · 03/05/2026

Paragraph 0063 in that patent is quite a statement. I haven’t figured out how to look at figures on the site, but that disclosure will need some rock solid data to be considered true. Of course, patents don’t necessarily need to be on real things that exist.

020

N-linked Glycosylation Enrichment for In-depth Cell Surface Proteomics of Diffuse Large B-cell Lymphoma Subtypes *

Agree the sites being on the cytoplasmic side are weirdest thing here. However, site is id’d in another one of unpublished Yang Mao’s n-glycoproteomes he did while he was still in CPH, and Mann lab published a site on TMPO www.mcponline.org/article/S153... so.. what if membrane topology is wrong?

mcponline.org

Global analysis of lymphoma genome integrity and transcriptomes tremendously advanced our understanding of their biology. Technological advances in mass spectrometry-based proteomics promise to comple...

E.g., proteins in nuclear pore complex (Pom152 in yeast)? These are the ones I think I remember being somewhat related to sialylation, but details are a bit fuzzy in my head!

000

Isn’t this paper talking about N-glyco on the inner nuclear membrane, but still pointing in to the lumenal space? Given the perinuclear lumenal space is contiguous with ER space, I could see things being synthesised in ER and being dragged along the membrane to the inner membrane?

310

The distribution of ConA-binding sites on oocyte nuclear envelopes

Given this paper claims oligomannose (or high man, haven’t checked!) it is less disturbing for me. Older paper also suggests ConA reactivity in nuclear envelope (www.sciencedirect.com/science/arti...). I would be curious to hear what ER / NM people think of the trafficking model?

sciencedirect.com

Ferritin-conjugated concanavalin A (ConA) was used as an electron-opaque tracer to study the distribution of oligosaccharides on nuclear envelopes iso…

100