fullsquare

their original promise was that: you will be able to plug a jar to a controller, download instructions and it will make medicine for you, no knowledge of chemistry needed. i think it was one talk before this one where they claimed to make an antiviral (still under patent) and gave away some pills of it for free. i don't think they made it in a way that matters, they didn't show their pathway, they didn't show analyses they say they made, they don't say what they used as a starting material and how they got it, and they didn't provide anything in way of formulation or other three drugs needed in that treatment (iirc). it might be very well that their costs are even higher per pill than what usual distribution will charge, but they don't say any of that, and it might be just as possible that if more people did that starting material supply would dry out. (or they might be just lying) worse than that, while they might do some analyses, it's not included or expected in jar and controller scenario, so there's no real quality control and if released, this alone can be reasonably expected to kill people. i'm not sure if they even released anything workable in this capacity (they also claim that they made pyrimethamine, but it's also after nilered made it, so they might just copy his process. it's simpler thing)

what they're doing later: there's this horse medicine that you can use as (more dangerous, less effective) ersatz plan B, or this estradiol thing, it's much less sketchy because they don't cook anything, they "just" formulate an API they get from somewhere else, but it's still a fair bit sketchier than using fish antibiotics in humans because for example massive accidental overdose is still quite likely

Think the issue is that some complexity can be removed without problem, and some absolutely cannot. And the problem of figuring out which is which is hard. (Which if you squint, seems to be similar to the chemistry stuff you describe here).

Well, i'm not exactly sure about it, but what i can do is describe how this process works in terms of operations and you can draw your own conclusions. There's not that much complexity in peptides in the first place, because synthetically, all you have to do is to make a lot of amide bonds, and this is a solved problem. Slightly bigger problem is to make it in controlled way, which is reason why protecting groups are used, but this is also a thing that has been around for decades.

The trick is to bind the thing you want to get to resin, which makes it always insoluble and therefore your product always stays in reactor. This can be an actual dedicated automated reactor or a syringe with a filter. We start with

Resin-NH-Fmoc

Fmoc is a protecting group that falls off when flushed with a base, so we do so, wash resin, and get

Resin-NH2

Then we can add coupling agent and protected aminoacid, for example leucine, then wash again, and this gets us

Resin-NH-Leu-Fmoc

then repeat. All operations are add reagent or wash solvent, stir, wait, drain, repeat. Deprotection, solvent, coupling, solvent, repeat. It's all very amenable to automation and it was explicitly designed this way. When all is done, resin is treated with acid which releases peptide, and because resin can be washed there are no leftover reagents.

Of course it can be all done by hand, and this allows for doing things like putting a couple of aminoacids on resin on a big scale, then splitting it into a couple of batches and attaching different things on top of that in parallel. Machine can't do this (at least machine like we have). Machine can instead run all of this while hot and this makes it fast, but sometimes things break this way, and also machine can run unattended for more than one shift (when it's not broken). Sometimes things fail to work anyway and it's a job of specialist to figure it out and unfuck it up. Sometimes peptide folds on itself in such a way that -NH2 end is hidden inside and next residue can't be attached. This can be fixed by gluing two aminoacids in flask and then using a pair instead of a single one in machine, bypassing that problematic step. Or in a couple of other different ways, and picking the right one requires knowing what are you doing.

Solution phase synthesis looks different because every step requires purification after it, which is sometimes a thing you can wing and sometimes not. The advantage is that when you need lots of product, you can just use bigger flasks, while bigger machine (and large amounts of resin) gets prohibitively expensive. Ozempic was made in solution (at least once) for example. Again, doing things by hand gets you extra flexibility, because machine can only make peptide from start to finish in single run, but if it's done in solution instead, you can start from, say, five points and then put pieces together (which starts to look like convergent synthesis, and this also makes it better for large scale). Machine can't do that (unless these pieces are provided, but at this point most of the work is done)

Looking back at these people, even when operations are simplified, there's no deskilling of operators that they aimed for, it's just throughput that increases. They also don't have the benefit of that "keeping the important things always in reactor" thing

i can't find that essay now, but i think it was written in latex, and also complained on top of cobol about java, ada (in military context) and a kind of non-programming where block diagram made by non-programmers was turned into executable

He wrote also negatively about the hackers who do homemade meds thing.

i've heard about them before and got reminded of their existence against my will recently. (do you know that somebody made a recommendation engine for peertube? can you guess which CCC talk from last winter was on top of pile in their example?)

you know, i think they have a bit of that techbro urge to turn every human activity into series of marketable ESP32 IOT-enabled widgets, except that they don't do that to woo VCs, they say they do that for betterment of humanity, but i think they're doing it for clout. because lemmy has only communist programmers and no one else, not much later i stumbled upon an essay on how trying to make programming languages easier in some ways is doomed to fail, because the task of programming itself is complex and much more than just writing code, and if you try, you get monstrosities like COBOL. i'm not in IT but it seems to me that this take is more common among fans of C and has little overlap with type of techbros from above.

so in some way, they are trying to cobolify backyard chemistry. the thing that is stupid about it is that it has been done before, and it's a very useful tool, and also it does something completely opposite than what they wanted to do. it's called solid phase peptide synthesis, and it replaces synthetic process that previously has been used in liquid phase (that is, like you do usually in normal solutions in normal flasks). (there's also a way to make synthetic DNA/RNA in similar way. both have a limitation that only a certain number of aminoacids/bases is actually practical). the thing about SPPS is that it can be automated, and you can just type in sequence of a peptide you want to get, and machine handles part of the rest.

what you gotta give it to them is that automated synthesis allows for a rapid output of many compounds. but it's also hideously expensive, uses equally expensive reagents, and requires constant attention and maintenance from two, ideally more, highly trained professionals in order to keep it running, and even then syntheses still can fail. in order to figure out what got wrong you need to use analytical equipment that costs about as much as that first machine, and then you have to unfuck up that failed synthesis in the first place, which is something that non-chemist won't be able to do. and even when everything goes right, product is still dirty and has to be purified using some other equipment. and even when it works, scaleup requires completely different approach (the older one) because it just doesn't scale well above early phase research amounts.

what i meant to say is that while automation of this kind is good because it allows humans to skip mind-numbingly repetitive steps (and allows to focus on "the everything else" aspect of research, like experiment planning, or parts of synthesis that machine can't do - which tend to be harder and so more rewarding problems) this all absolutely does not lead to deskilling of synthesis like this bozo in camo vest wanted to, i'd say it's exactly the opposite. there's also the entire aspect of how they don't do analysis or purification of anything, and this alone i guess will kill people at some point

Derek Lowe comes in with another sneer at techbro-optimism of collection of AI startup talking points wearing skins of people saying that definitely all medicine is solved, just throw more compute at it https://www.science.org/content/blog-post/end-disease (it's two weeks old, but it's not like any of you read him regularly). more relevantly he also links all his previous writing on this topic, starting with 2007 piece about techbros wondering why didn't anyone brought SV Disruption™ to pharma: https://www.science.org/content/blog-post/andy-grove-rich-famous-smart-and-wrong

interesting to see that he reaches some of pretty much compsci-flavoured conclusions despite not having compsci background. still not exactly there yet as he leaves some possibility of AGI

damn i see that chatbots don't want to stay behind rfk jr in body count

will they learn that safety regulations are written in blood? who am i kidding, that's not their blood

firefox at minimum clears the very low bar of not exposing casual user to crypto

That might be why they do this, marketing is secondary (and mcd at first didn't market fries as tallow-containing specifically). When done on purpose and currently, i guess it's seed oils thing

painting goth David Gerard as their archenemy is pretty rich coming from a group that already owned two castles

Tallow is saturated, so you can use it longer before it oxidises

no