Connecting AI directly to the data sources (instead of just asking it to provide code that I run locally for myself) can get quite complicated in terms of meeting institutional policy, applicable law, data access-storage requirements (e.g. NIH data repositories), and can require legal agreements between institutions and the AI provider.

I cannot touch. At least not yet.

How do you validate this kind of work to weed out any confabulating by the LLMs?

I'd really like to see much better visualization from Claude Science at some point. Educational-esque, with full threejs + shaders scenes over just these plots and protein/chemical structures. This for a lot of papers in the literature review would be awesome.

Previously integrated Biomni into our intelligent workspace:

https://blog.codesolvent.com/2025/07/ai-assistant-with-biome...

Happy to chat if intrigued.

Sounds like the perfect use case for some kind of framework where you have a local LLM (that can run on lower spec hardware) collaborating with the main LLM to optimise latency and all the other niche and legacy use cases ?

I agree that it's an interesting architecture, but I'm not sure how it would work in a highly controlled server.

If you can't connect from your Mac, then I doubt they will allow an agent to make requests from the server

> Then Opus 4.8's safety system flagged the session.

If you think you can use this to land real positive impact, you, your institution, or your company should apply for OpenAI and Anthropic's bio programs!

Can you speak to what makes this different from simply including or configuring various agent skills? Or is it simply the combination of lots of helpful defaults that makes this product useful?

Anecdotally, as someone with a lot of moderately computational sciencey tasks at work (part of my job is as a data analyst for a geology firm that has some interesting sensor data), combining Claude Code and standard python data libraries has been extremely powerful and sped up my workflows immensely. If I just need a quick analysis or visualization, Claude can write something for me in minutes that would take me an hour or so to sort out on my own. I know the relevant libraries well enough to read and verify the code, which is an important distinction from blindly using a black box AI.

I will note that Claude Code and Jupyter in VSCode don't play nicely together right now - it forces me to rerun the whole notebook from the start after every edit Claude makes. This has led to me stepping back from notebooks and having Claude write standalone scripts that I then spend time merging back into a pretty notebook.

I can't speak for Claude Science, but I prefer using Biomni as an agent for bio over Claude Code with a custom setup because a) Biomni stays on the frontier for bio, b) it has a config that just works and skills I trust are correct, and c) it has better built-in abstractions for long-running sessions.

As a concrete example, computational biology jobs sometimes run for hours on the Biomni HPC. When they're done, the session needs to reawaken, process the results, iterate, etc. You can implement something like this with agent callbacks, but it's not as straightforward.

This repeats many times for many integrations, so it's just simpler for me to use an agent that's built for exploratory bio and already has all of this. Claude Science has some of these features, so I imagine they're aiming for something similar.

The FAQs at the bottom of the page answer your question.

> Then Opus 4.8's safety system flagged the session.

The jokes write themselves these days.

I suggest collecting 10 seminal works on the subject matter including 10 textbooks in the general field, converting them to plain text via OCR or text extraction, then trying the same thing with a superior agentic harness, like omp.sh

/goal set create biopesticide targeting the DvSnf7 transcript of western corn rootworm

<sarcasm>make no mistakes</sarcasm>

My friend they have dumped hundreds of billions into LLMS.

The ROIC is not gonna look good if they do not somehow make use of the existing assets...

Not an argument for btw, Im just saying. Ultimately the management answer to shareholders who look at return measures such as that.

Competition is healthy, yours looks cooler.

A lot of the soft and hard sciences use hacky matplotlib code to produce results and visualisation, without being necessarily data science

From the bits I've seen, I'd take claude-generated code any time over that written by maths, physics, biology, linguistics people. Even though I've seen Claude make some super-big mistakes while doing data analysis I'd guess it's already more reliable than most academics trying to code.

My take based on the video is that they're thinking more about bioinformatics, which might technically fall under the "data science" umbrella depending how you define your terms, but which is not described that way in common usage.

It's the content that determines the sort of science, not the toolchain.

All of these new things are starting to look like soviet space program propaganda. Is there something really new?

Well LLMs are largely useless and people are realizing that.

unfortunately no arch based distro support. I'm curious why it's not packaged as a flatpak.

They're gonna worsen it

In some fields like comp sci, when code isn't given but the paper describes the approach, LLMs do help with the reproducibility crisis: you can ask it to reproduce the result through reimplementation by reading the paper.

If it fails you may have to double check it did properly reimplement it, but if it succeeds you do get a reproduction.

lebovic answered this, but it isnt just claude cowork especially with connection and abilities related to SPC clusters. I could defientiyle see my former team at a national lab integrating this with their systems, and forgoing the use of Claude Code all together

This thing is also surprising considering Fable was not allowed to answer any biology questions.

I think presentation via software just isn't a lot of their strong suits. A lot of researchers' personal or research lab sites too are usually way out of date or just really badly presented from what I've seen. They could all do with some thinking about aesthetics and understandability more.

Conveniently, you can use published results as tests of equivalence, provide the ugly code as context, and regenerate it to your liking. I think the odds of such a regeneration introducing a bug that's within the usage domain but that dodges the golden tests are quite low... so long as you resist the urge to add features along the way.

Matplotlib? Ha! There are loads of academic fields where you still write data analyses by hand, one at a time, in Matlab, without proper version-control or libraries.

It's not obvious from the marketing if this is applicable to non-biosciences. If it is, they couldn't come up with a single example from another domain like astrophysics?

https://www.anthropic.com/news/claude-science-ai-workbench

EDIT: Installed the app, it has zero connectors for non biology, which is a shame. I assume they'll come later.

Old wine, new bottle...

The thing that really scared me about the landing site for Claude Science was this promotional image of the software in action:

https://cdn.prod.website-files.com/6889473510b50328dbb70ae6/...

Very depressing. MDPI journals will be saturated with these slop papers (if they're not already). It shocks me that Anthropic thought that this was a good thing, and says a lot about their research integrity (or lack thereof).

> Science has no idea yet how such disclosures should work yet.

Technically, most journals have a policy that LLM use should be acknowledged, but I agree we're still very much in the weeds about this right now. Much firmer guidelines should have been established years ago.

(I also have no issues with LLM usage in research either, btw -- I use LLMs to fact-check / proofread / discuss / sanity-check my conceptual work, to background myself in other research, and to refactor and assist with analytical coding. They can be a game-changer for medical research, when used rationally and sensibly.)

pharma is currently in a tailspin and not really spending money. they'd rather outsource everything possible to china or india right now.

First party support would be nice since this is not a high-trust in the AUR period, but fair point, I'll probably use it, thank you!

We send our regards: https://arxiv.org/abs/2510.15061 (ICLR 2026)

They do mention things like protein and chemical structure visualization though

Oh it's getting there. I've turned down several referee requests this year because the paper looks like AI slop. A lot of it seems to come from China.

Isn't this just blanket cynicism?

In the long run conceivable we could use AI to hold papers to a much higher standard, audit all the data and code that is associated etc.

No its not doing magic. Im impressed when anyone can play a guitar, because I dont know anything about playing a guitar. Someone who's been playing the guitar for years isnt impressed by all guitar players.

This seems the case with many people using llms to write code. They think everything an llm does is magical.

It will never be able to replace humans with two brain cells.

This plus it's entirely plausible their employees have access to Fable or their own other pre-released models internally. Other than the perks you mentioned they've got excellent distribution too.

Most of what's impressed me in working with LLMs is just how much "intelligence" you can get out of the agent iteratively refining something it looks back at with each turn, without its ever actually exhibiting human-level intelligence. I've always been an embodied-cognition guy, and it really seems to me like "agent harnesses" are basically task-specific pseudo-embodiments for LLMs.

Underlying reproducibility is integrity.

Underlying integrity is rigor.

Underlying rigor is education.

It goes deep, for sure, IMO.

Some authors may even choose to leave syntactical errors as a tell for those self-authored passages; long-term, some interesting language drifts may come of it.

now try it again.

Its kinda both. Its quite underwhelming at the top, but at the bottom its amazing.

Tell us, what gives you that impression?

Eli Lilly's recently partnered with NVIDIA to spend a lot of money for a new research lab in the Bay Area so not entirely

Perhaps it is true that the faster you can internalize knowledge (thoroughly, there is a quality aspect to it), the faster you are. Maybe I'm getting old and learning new skills is getting tougher. Maybe, as my world model grows I'm becoming a slow thinker, or a slow learner? New stuff has to be evaluated against a lot of knowledge. But when it clicks, it really feels like a click, it feels satisfying. Like when some new knowledge does not just explain the problem at hand, but also a lot of things that still lingered in the back of your mind.

Recently my wife said that my daughter (ill at the time) may have heatstroke, my response was: It looks like it but she also has a hefty fever (hot after being more than 24 hours out of the sun), I can't really imagine the immune system being involved in heat stroke, although it's possible... My mind went out to heat damaged proteins presenting neo-antigens triggering an immune reaction. I also labelled that as unlikely and more dangerous than what we were observing. I like that I can do that (of course I went on to verify these thoughts!). That reasoning, it's not exactly 100's tokens a sec, but I like the process and it has value.

I also recently observed some weirdness in a dataset, I spend 3 days hunting it down. Long story short: I though I understood how genes make transcripts but I was wrong and ended up adding a new transcript to the human reference genome annotation together with the Gencode people. Now I understand my data better and can separate two different transcripts better in my data (a difference important to our research).

Things like that. The LLM doesn't speed that up, not really. I read a part of a book on gene expression and the function of transcription factors and their interaction with promoters, but I also used LLMs, In the end it was the book with the pictures and clear language that communicated the concepts most clearly. It was made for that of course, and I knew I could trust it (it's tiring to assign <100% confidence to LLM answers), although I know a real scientist also does that with books :)

Maybe I, we all (humanity), will really be faster in the future. Maybe when you grow up with these things you can build world models better and faster. Maybe I'm just too stuck in my ways, as my neuro-plasticity degrades over time. Or maybe it doesn't degrade, maybe I just need more evidence before changing my world models, they have been building on a heavy foundation for a while now.

Not sure how to feel about this. I think its super cool that you can dive into this, but it sucks that its your son that has this condition for which you have to do this analysis. I hope it all turns out well.

Quick question: where did you get your genome read and get the raw files? As far as I know, as service like 23andme does not give you back the raw files.

You're not worried your whole genome is being sent over to some commercial entity?

> audit all the data and code that is associated

For a while now there has been very little incentive for providing these alongside the paper, and I don't see why exactly 'AI' would change this. I could even see how making it vague to be harder to test with LLMs could be profitable for citation hackers.

Unless reviewing becomes more profitable than publishing, anything that makes both easier will drive one up far more than the other. And it is difficult to conceive of something that would make reviewing much easier without making publishing much easier.

Lily is probably the one major pharma company doing alright these days. That being said this thread has a more sober take:

https://old.reddit.com/r/biotech/comments/1rgjnrj/lilly_bets...

I don't hold that view exactly. But something related...

I once tried to replicate a bioinformatics result based on published data (for a class). I found that although the process did indeed yield plots A and B, as the authors claimed, they were typeset wrong in the PDF so plot A had B's caption and plot B had A's caption.

It would be an easy thing to provide assurances against, if you wanted to. You could repeatably build the pdf so that such a mistake was in plain view, as a bug in the pipeline, rather than something you had to do offline calculations to support or reject.

The situation as it is is not ideal. Instead of anything that would verify either side, it's my word against the author's until a third party bothers to repeat the analysis. That's the best we can do for scientific claims, but there are friendlier ways to make the computational claims verifiable.

The Claude science video showed a little "provenance" button and talked about exactly this. Life sciences have their hands full with the actual science. They're not immature, but they are not in a great position to be proving the validity of the computational connective tissue that underlies their results. That's a whole thing on its own, independent of the underlying scientific reasoning being presented (though I wouldn't call it data science).

Plus, its exactly the sort of thing we need AI to get better at: sourcing evidence that proves its claims and stitching it together so the proof is easily verifiable.

I too am excited.

You can imagine using AI agents to tag papers that don’t have code or similar work attached and just filtering them out.

The Chinese open source community has made a lot of incentive to make research reproducible for example. The most reproducible works from I.e. deepseek get widely cited and adopted.

I don’t think we can just say “AI” and it’s fixed but with deliberate effort there’s reason to be optimistic.

Marginally but the data manipulation is actually being done locally as the genome CRAM files are like 24Gb each.

Your comments really resonate with me. I have been trying to square the rapid progress in AI's abilities to generate proofs of unsolved math problems with comments that Terence Tao has made regarding whether or not these (incredibly impressive) models are actually contributing to mathematical progress.

From Tao on Mathstodon on April 27, 2026: "We are transitioning in mathematics from an era of proof scarcity to an era of proof abundance, but our mathematical infrastructure and culture has not yet adapted to this. As mentioned previously, there is now a strong (and growing) impedance mismatch between the three core components of mathematical problem solving: proof generation, proof verification, and proof digestion.... Perhaps surprisingly, this massive acceleration in proof generation has not actually produced significant acceleration in mathematical progress itself (with the possible exception of #1196, in which all three stages are largely carried out at this point, and for which some digested assessment of developments will soon be forthcoming)."[0]

And I guess my question is, will it matter if humans grasp these new discoveries? If the models are capable of incorporating these discoveries and using them to recursively self improve to unlock new discoveries without humans in the loop, then I guess we humans never really need to understand. I find it hard to believe that a machine can understand concepts that we never will, but I can't reason why that couldn't be the case. Something that holds trillions of concepts in its mind at once might be capable of generating a proof to something that we simply aren't capable of understanding. And the machine just tells us, "Listen, if you are too stupid to understand the new laws of physics I am giving you, then simply follow these very explicit instructions on blasting hydrogen with this laser at this angle in this exact magnet conformation, then you will get cold fusion."

I am really struggling with this. I think superintelligence implies that there will be things about the world that the models understand that we won't. And I can't quite articulate why that is depressing. Because we should still get some cool new tech and some life saving drugs.

[0] https://mathstodon.xyz/@tao/116477352332170731

it's not, the genome is treated locally by tools called by the LLM, the LLM itself can't do much with the raw DNA sequence

I wouldn't care where my data went if it was to help my children.

Just as a counterpoint ML and AI research has become much more reproducible over time. I feel like this is relevant because ML / AI researchers are huge power users of AI tools.

Between 2016 and 2021 the share of ML/ robotics/ AI researchers being reproducible (ie contianing code and similar instructions to reproduce) doubled [1].

The major US labs have gone largely closed source (I.e. they no longer publish frontier research) but the Chinese ecosystem has incredibly reproducible code.

This is field dependent obviously but I think it atleast gives reason to be optimistic.

Yes people will churn out fake slop research, but it feels like that can be categorized and then ignored.

[1] https://arxiv.org/pdf/2308.10008

Thanks for the kind words. Actually we got the trio whole genome sequencing through our neurologist/geneticist a couple of years ago. It was performed by a company called GeneDx. They interpreted the data at the time which is how we got to a diagnosis but knowing I'd want to dive in to it later I asked for the raw data. They provided it as the raw CRAM files and also the VCF (variant call files) which are a bit smaller. But each company has its own pipeline and for example uses different versions of the reference human genome which made working with the data quite hard for me and the people I enlisted. Claude Science seemed to make very easy work of it. Also to be clear, the question I was trying to answer was whether his mutation was likely passed down through my sperm or mom's egg - neither of us have the mutation in our own genomes. Turns out spontaneous (de novo) mutations are much more common in sperm because Spermatogonial stem cells have undergone many more cell divisions over their lives. Everyone has de novo mutations (70+), one of his just happens to be in an unlucky location.

GeneDx aren't direct to consumer so you'd need to get it ordered through a physician but there are some DTC options for example, Dante Labs, Nebula Genomics, Sequencing.com but I can't speak to the quality of their testing.

23andMe doesn't do whole genome or whole exome sequencing. They use a microarray technology that tests for about 650,000 single nucleotide polymorphisms. You can actually download the your raw data on 23andMe and do your own analysis or use a tool like promethease.

I'm an MD so I'm quite comfortable exploring this data and whatever it uncovers. Tools like Claude Science are going to put a lot of power in the hands of every day people, potentially outside the guidance of genetic counseling/docs, which many organizations in the past (including the FDA) have been hesitant to allow.

That's good to hear about ML and AI research, but most research is not based on computers and so would require laboratory setups to reproduce. Not only is trying to reproduce such findings (beyond what is effectively a sanity check) through simulations a lost cause, if AI can reproduce such research it would be capable of doing such research itself... in which case it would be far more fruitful to use AI to do further research.

I can vouch for Nebula. Just follow their recommendation to not get the 1x service if you want to use the data for medical reasons. The more passes they make, the more accurate the result.

Downside: those files are HUUUUUGE. Have a good reason to do it before pulling one down and trying to work with it.

This thread is about a product based on fully reproducible research though so I feel like we should stay grounded. Claude science is meant to be used in the context of reproducible science research, there is a decent reason to not be cynical on future research being reproducible.

> if AI can reproduce such research it would be capable of doing such research itself

Well there is a big distinction between research validation and research generation, it is generally much easier to verify that a math proof is true or false than to find a truly novel proof.

But yes in the long run I’d think AI will be doing tons of research and it will by default reproducible. So maybe we’re aligned after all?

Files are huge for sure (mine are like 25Gb x 3 of us) but Claude Science / Macbook Pro M5 Max knocked out my analyses in minutes using command line tools like bcftools and scripting. Honestly think this opens up WGS data to much easier analysis and refreshing those analyses in minutes as new discoveries are made.