koeng

We’re pretty close. There’s TEM microscopy tech which basically tilts a sample to get a bunch of lines, which is then reconstructed as a 3d model.

It’s stupid expensive though, and you can only really identify whole proteins. But you can do that in context, which is massive

I'd love to try it, but pretty much all my repos are >10mb. It's not because there is that much code, but because I am doing bioinformatics and the test files (for the unit tests) inflate the repo size. It would be great if there was a way to test it on just 1 large repo for perhaps a week or something, because I balk at the idea of spending $20 a month on something that I don't even know works well.

This is important because I'm not deeply familiar with public projects, so I can't accurately assess if the tool is worthwhile. Whereas with one of my repos, I'd be able to tell quality pretty quickly.

Is there an API that could be used? I have a use case that I'm integrating into a larger software package, but wouldn't be using a cli/web app for that.

I use hledger for my solo-company. It is really great, and very easy to understand what is going on. Highly recommend.

> The big issue seems to be the size of most plastics, not that bacteria can't already eat them.

Yep, that's why it is super cool things like mealworms can consume styrofoam - they have microbes in there gut that do the actual digesting, but they can chew it up physically, which is very important.

Back in ~2014-2015 I was looking at making this (worked in Chang Liu's orthorep lab) - the problem was that I couldn't get my hands on the PRD1 phage (which is a fascinating phage - protein primed replication, lipid membrane coliphage). They were clever about how they bootstrapped the system - they just expressed a whole lot of terminal protein and hoped that some would attach. And it did in VERY rare events.

The most annoying part with the 200k polymerase prize is that they don't actually provide protocols for validating the polymerase works well. Me and some friends can definitely do the AI-protein design + DNA synthesis + protein synthesis (and would want to), but no way am I gonna figure out idiosyncratic sequencer protocols.

Oxford Nanopore with the new R10 cells is good enough for most synthetic biology applications (ie, plasmid sequencing and strain validation), which I'm super excited about. Combine low-capital and long-reads and you can do some really interesting things in the DNA assembly realm. It's pretty recent they got there though

> At some point us biologists only have ourselves to blame for our purchasing decisions and their consequences

Yep...

Yep. You can also do things like mRNA vaccines, or build industrially useful enzymes, or make crops resistant to disease. Pretty much any biotechnology that involves engineering life.

How can we lower the price of making synthetic DNA?

All the graphs on this topic are out-dated, specifically because the hype-train doesn't want to acknowledge that the price of usable synthetic DNA has gone up - not down (even when balanced for inflation) - over the last 5 years.

Enzymatic DNA synthesis companies are not fixing this, or if they are, are at least 5-10 years out. The fundamentals of DNA synthesis are actually super cost effective. The raw price of a 100bp ssDNA vs an assembled gene, ready to be used in a biological context, is ~200x. Even at small-ish scale, the price for a full assembled gene from the same supplier is $125 per kilobase, while the small synthesized fragment is only $4. And they're STILL losing money on the assembled genes.

Assembling genes is really unsexy, though, so the field is kinda stuck in hyping DNA synthesis while ignoring fundamentals of DNA assembly.

Their media is real complicated and expensive (plus difficult to prepare: can’t just throw everything in an autoclave), and they don’t grow very fast, and they don’t grow to a high concentration. Basically that means that you can easily get contamination (like you would growing human cells), and even when you do get growth, it’s dilute.

I’m well aware. I was in charge of designing and synthesizing the entire gene set, optimized for E.coli, in order for scientists to easily try to discover the function of these genes https://stanford.freegenes.org/collections/gene-sets/product...

Out of the 100 or so unknown genes we had/have, only like 3 unknowns were actually used for research. Such is science sometimes.

I gave a talk at that workshop! (I’m making the 61 codon version of JCVI-Syn3a)

You can get the cells as a researcher, but basically forget about it for a commercial user or hobbyist. They’re also annoying as hell to grow. I’m trying to synthesize it from scratch for the precise reason of freeing it from MTAs (material transfer agreements)

We don’t even know what like, 20% of the genes do. Simulations are neat, but they miss a lot of underlying biology that actually makes an impact on engineering.

One of my favorite papers was published 40 years ago on this topic https://pitp.ias.edu/sites/pitp/files/morowitz-completeness_...

I love your demo for this. It's one of the best demos I've ever come across in a launch HN. Very easy to understand and use. It seems to suffer with more complex questions though. For example:

Question: Why does the pUC19 plasmid have a high copy number in bacterial cells?

Expected answer: The pUC19 plasmid has a high copy number due to the lack of the rop gene and a single point mutation in the origin of replication (ori) derived from the plasmid pMB1.

GPT response: The pUC19 plasmid has a high copy number in bacterial cells due to the presence of the pUC origin of replication, which allows for efficient and rapid replication of the plasmid.

Both are technically correct - the expected answer is simply more detailed about the pUC origin, but both would be considered correct. It seems difficult to test things like this, but maybe that's just not possible to really get correct.

I wonder how well things like FutureHouse's wikicrow will work for summarizing knowledge better - https://www.futurehouse.org/wikicrow - and how that could be benchmarked against Talc

Fair on the gatekeeping, but on the other side, keep your mind too open and your brain will fall out.

> This is precisely the problem I’m pointing out.

This is, ironically, also the problem I’m pointing out. Why would I trust someone who doesn’t work in or on computers in computer security?

Arguing that once the promise is shown it’ll become super dangerous is essentially unfalsifiable. The same argument can be made for literally any new technology. I’d like to see reasonable chains of logic.

Eh. Most people I see who are super concerned about synthetic biology safety aren't the people actually working on it (in contrast to things like AI, or even more innocuous technology like social media). That includes me. I've worked in synbio for >10yr and consistently the danger is overhyped by people who don't get it.

> dire, irreversible consequences we can't even begin to predict them

No, they're dire consequences that you can't predict, probably because you don't work in the field. There are lots of ways to predict what could happen.

Not a ginkgo project

Hey, a project I actually know about. Used to work in Kaisha Benjamin's lab (Endy Lab at Stanford). I remember all those bioinformatics to get the glucose inducible promoter. Back then at least, the promoter wasn't controllable enough. Currently, I'm actually vaguely helping with this project by synthesizing phage PH15 (building a high throughput DNA synthesis/assembly company right now). Can probably use the phage to package up DNA to transform S. epidermidis.

In general, I think microbiome engineering is super cool and should be explored more!

Sex changes many lines at a time. But you do die if it doesn’t compile.

I’ve worked in a lab of one of the original folks working on genetic logic gates, and honestly, no they don’t really work very well at all.

I think a lot of computer scientists assume that their model of the world (binary logic) is the most abstracted version possible, and you can drill down other information carrying systems to that level, and then use that knowledge to build reliable systems.

Biology does not work that way. Its fundamental abstraction is different (mostly one of massive interconnectedness). Engineering logic gates and the like, mostly, doesn’t actually allow you to build better genetic circuits.

Can you explain why the streamed WAL files are a copy and not a backup of the SQLite database?

The broader security mindset is identifying the technologies that are most probably used in a situation where you need read and write. For example, in computer security, you wouldn't worry about a air-gapped UNIX server being infected by an online windows worm. Undetected write and execution are not enabled by RT, like, at all. Synthesis (write) is pretty much a DNA technology, and the only time that a RT is used is for proper viral packaging (lentiviral transformation, mostly). But there is also not really a need to obstruficate because there are tons of completely valid uses for lentriviral vectors. In fact, there are open source RT plasmids available because they can also be used for things like production of open source COVID tests (for example, during COVID I deposited this expression vector under open terms https://www.addgene.org/165556/ ). If you really want to get into the weeds, there are retrotranscriptases that are naturally expressed in humans: good luck with false positive hits on detection.

More broadly, in biology, there isn't really a thing as undetectable execution, because there are physical molecules floating around, and furthermore there are trillions of natural self-executing blobs in an arms race with the ribosomal world. Detecting execution, as technology gets better, is more of a problem of privacy than technology.

Undetected write is more important, but it is also important to identify the particular technologies that actually matter. It's not RTs, nor is it Cas9, nor is a lot of the things that people typically point out on hackernews, either.

From someone who has reconsistuted viruses from synthetic DNA - it’s WAY easier than working with prions, or even viroids for that matter. The fact they package themselves and release themselves, while being formed of genetic sequence, is SUPER nice. Any bio computer doing sequence manipulation will probably find it way easier to use a virus-like replicating “malware”

I don’t see how that has to do with obfuscating unusual reverse transcriptases, though. There are plenty of (if not more) DNA based transformation methods, and those enzymes are sufficiently widespread that it’s not unusual to find them nor is it particularly hard because of the structures

> * A possible arms race between obfuscating and detecting unusual retrotranscriptases

Why, exactly?

> Wonder how badly it could go if the data storage container is ruptured while against someone's body?

Nothing, really. Your body is good at fighting that sort of thing, because it's literally happening every second of every day with billions of bacteria and environmental DNA. Even specialized organisms have a rather tough time getting by all the defenses.

For example, there are zero human-infecting raw-DNA pathogens.

How would you do it in Go? It is my understanding that FAISS is integrated into the SQL database itself (using embedding in the table), so writing it in Go wouldn't integrate it in the same way.

> But you can try to use my build scripts, and tweak my code, to build SQLite+sqlite-vss+Faiss into one large WASM

That makes perfect sense to me! I'm looking into your build scripts right now. I'm pretty proficient in Go, but haven't touched wasm or C much, so. I may reach out with more questions.

Very cool! I’ve been looking at using SQLite-vss for fast vector sorting, but it requires CGO. I imagine if I do WASM compilation directly from C, I might be able to get around using CGO. Do you have any idea able the feasibility of that using your package? Thanks!

One thing that I've been doing lately is creating a "synbiogpt", and from it, have come to realize the limitations of the custom GPTs.

- Biological sequence data is usually quite long. This is fine if the biological data is in a file: however, if you need interact with an API for advanced function (like codon optimization), you have to send this across a wire. The API calling context window then gets filled up with sequence data, and fails.

- I can't inject dependencies, many of which I've written myself specifically for biological engineering. Sometimes GPT will then try to code its own implementation, often which is incorrect.

- The retrieval API often fails to open files if GPT-4 thinks it knows what it is talking about. When I'm talking about genetic parts, I often want to be very specific about the particular parts in my library, rather than the parts GPT-4 thinks is out there.

I fixed most of this by just rolling my own lua-scripting environment (my biological functions are in golang, and I run gopher-lua to run the lua environment). I inject example lua for how to use the scripting functions, as well as my (right now, small) genetic part library, and then ask it to generate me lua to do certain operations on the files provided, without GPT-4 ever looking at the files. My internal golang app then executed the scripted lua. This works great, and is much faster than a custom GPT.

The biggest problem I have right now is the frontend bits. I would love to have basically an open source ChatGPT looking-clone that I can just pull attachments out of + modify the initial user inputs (to add my lua examples and such). So far I haven't found a good option.