Chris Rands

@adamlewisgreen @shae_mcl Do you think training on sequencing reads rather than say counts matrixes or VCF will be the winning strategy? Does SRA contain enough annotations and metadata to be useful?

We already have PDB for biology, it's called SRA (and unlike PDB, it's ~100 petabases). Asking for all the omics in every possible combination of cell type, state, patient etc. to learn the "grammar of gene regulation" is analogous to demanding the internet contained every possible human conversation with accompanying parse trees before training GPT-1. Improvements in assay technology are certainly welcome, and we would not have SRA without historical improvements in tools for reading biological state; but on the current margin the bottleneck is implementing scalable unsupervised pretraining on existing cellular dynamics data, of which there are a surfeit. As we demonstrated in our recent paper, when you do this, many of the things which were thought to require specialized data (e.g. regulatory grammar can only be learned from functional genomics data; perturbation prediction requires mountains of exogenous gene knockout data) begin to fall out for free, exactly as they did in NLP. These specialized data sources will no doubt be useful, but as the icing and cherry on top of the unsupervised pretraining cake, not the cake itself. Some PDB and SRA back-of-the-envelope calculations from our 2024 work:

It’s estimated that the Protein Data Bank (PDB) cost around $13B to create. Alphafold was only possible because of it. If we want ML to solve biology, we should be funding the creation of databases and the development of new assay technologies. ML is nothing without data.

@biocheMichael @Bfaviero @grok Keytruda is nearer 30B. But I think this misses the point. Tech, financial services and luxury are the best businesses but pharma is in the next tier and really benefits humanity (or PR if you’re cynical)

@Bfaviero @grok Humira and GLP1 peptides are the best sellers of all time. Humira is like 20B a year and similar for GLP1s. Regeneron has several blockbuster drugs which sell 5-10B a year.

It makes no sense to me why OpenAI or Anthropic would make drugs Drugs which are ~wildly~ successful generate $1B in annual revenue Wegovy is not the norm and even if it was it took >40< years from discovery to Wegovy Sam Altman was 1 year old when GLP1 was discovered. Seriously think about that for a second. OpenAI burns $1B (now - pre drug discovery) in less than 1 month and they’re still scaling 10x or more… No fkn way are they going to stand up discovery through CMC and manufacturing and also clinical, reg, and other departments to support this fantasy Turn on ads and boom $100B a year in revenue Turn on drug discovery and oh shit it’s another ten years before we even see $1B in revenue. By then they could be a $10T company with $500B in ad and other revenues a year Someone explain it to me like I’m a drug discovery scientist

@morgancheatham Not unique to the genome at all. E.g. predicting novel features from pathology images like H&E

one of the reasons i became a medical geneticist: the genome is the only medical test where we measure once, but our interpretation evolves indefinitely. as our models and variant knowledge mature, the same data yields new truths, and eventually, actionabilities. the genome (and its derivative products) are the ultimate substrates for AI in medicine.

@JohnCarreyrou Big fan of your work John and Bad blood in particular, but I don’t see any new evidence presented here or a smoking gun. Also what is your response to the different C/C++ coding styles of Satoshi and Adam?

If you didn’t participate in the discussion the Bitcoin white paper sparked on the Cryptography list in late 2008, why would you tell a podcast you did 5 years later? Why lie about that?

@lpachter Let’s port the whole of bioconductor and be done with R!

I used Claude Opus 4.5/4.6 (and a bit of Codex GPT-5.3) to port edgeR to Python. See edgePython github.com/pachterlab/edg… This allowed me to develop a single-cell DE method that extends NEBULA with edgeR Empirical Bayes. All in one week. Details in doi.org/10.64898/2026.…

@Ronalfa Should I not bother removing the asbestos from the house then?

Okay we're going to cure all human disease in some unreasonably small quantity of time. Wire instructions to follow.

@dr_alphalyrae As GLP-1 drugs spread, the bell curve of humanity narrows- less obesity, addiction, depression, obsession, neurodivergence, innovation, creativity

Is anyone thinking and writing about how the proliferation of GLP1 agonists will change human society and culture? I genuinely think that this might happen

@AcquiredFM PepsiCo?

Tomorrow.

@ValthosTech @grok unpack this for me with concrete example , be specific

Valthos builds next-generation biodefense. Of all AI applications, biotechnology has the highest upside and most catastrophic downside. Heroes at the frontlines of biodefense are working every day to protect the world against the worst case. But the pace of biotech is against them: more powerful methods to design biological systems, with near-universal access, open up an increasing surface area of threats. In this new world, the only way forward is to be faster. So we set out to build the tech stack for biodefense. Our team of computational biologists and software engineers applies frontier AI to identify biological threats and update medical countermeasures in real-time. We are backed by $30M from @OpenAI, @Lux_Capital, @foundersfund and others including @Definition_Cap. We are actively hiring engineers to join in the mission - if that sounds like you, get in touch.

@AcquiredFM Please, no spoilers!

@c_rands Essentially. So much of this story is the battle against Microsoft.

Tomorrow.

@PritsWire @sama Fork handles??

@c_rands @sama FOUR CANDLES!

if you are a power user, please send us feature requests! (i asked in reply to this message and they were interesting, so would like more)

@PritsWire @sama Fork handles?

@c_rands @sama And four candles

@andrewwhite01 Is the logical conclusion here that papers become like a JSON with MCP servers?

I've written up some thoughts on publishing for machines. 10M research papers are published per year and there are 227M total - machines will be primary producers and readers of publications going forward. It's time to revise the scientific paper.

@srikosuri But many of the major causes of death like oncology, cardiovascular and Alzheimer’s are correlated with ageing? Also birth rate declining feels significant?

The exceptions to this: 1. Aging 2. Cosmetics 3. Performance Enhancement 4. Recreational I think that's it.

Pharma/Biotech is a negative sum game. The better the sector is at doing its job, the smaller its market gets over time.

@slavov_n Unfortunately humans are a little more complex than fruit flies

We should always strive to acknowledge problems and improve. We should also acknowledge when things work. 𝐘𝐞𝐬, 𝐦𝐚𝐧𝐲 𝐫𝐞𝐬𝐞𝐚𝐫𝐜𝐡 𝐫𝐞𝐬𝐮𝐥𝐭𝐬 𝐚𝐫𝐞 𝐫𝐞𝐥𝐢𝐚𝐛𝐥𝐞 & 𝐫𝐞𝐩𝐫𝐨𝐝𝐮𝐜𝐢𝐛𝐥𝐞. nature.com/articles/d4158…

@random_walker @sayashk @grok what do you think? Time to retire?

Some aspects of AI discourse seem to come from a different planet, oblivious to basic realities on Earth. AI for science is one such area. In this new essay, @sayashk and I argue that visions of accelerating science through AI should be considered unserious if they don't confront the production-progress paradox. ========= AI leaders have predicted that it will enable dramatic scientific progress: curing cancer, doubling the human lifespan, colonizing space, and achieving a century of progress in the next decade. Given the cuts to federal funding for science in the U.S., the timing seems perfect, as AI could replace the need for a large scientific workforce. It’s a common-sense view, at least among technologists, that AI will speed science greatly as it gets adopted in every part of the scientific pipeline. But many early common-sense predictions about the impact of a new technology on an existing institution proved badly wrong. The Catholic Church welcomed the printing press as a way of solidifying its authority by printing Bibles. The early days of social media led to wide-eyed optimism about the spread of democracy worldwide following the Arab Spring. Similarly, the impact of AI on science could be counterintuitive. Even if individual scientists benefit from adopting AI, it doesn’t mean science as a whole will benefit. When thinking about the macro effects, we are dealing with a complex system with emergent properties. That system behaves in surprising ways because it is not a market. It is better than markets at some things, like rewarding truth, but worse at others, such as reacting to technological shocks. So far, on balance, AI has been an unhealthy shock to science, stretching many of its processes to the breaking point. Any serious attempt to forecast the impact of AI on science must confront the production-progress paradox. The rate of publication of scientific papers has been growing exponentially, increasing 500 fold between 1900 and 2015. But actual progress, by any available measure, has been constant or even slowing. So we must ask how AI is impacting, and will impact, the factors that have led to this disconnect. Our analysis in this essay suggests that AI is likely to worsen the gap. This may not be true in all scientific fields, and it is certainly not a foregone conclusion. By carefully and urgently taking actions such as those we suggest, it may be possible to reverse course. Unfortunately, AI companies, science funders, and policy makers all seem oblivious to what the actual bottlenecks to scientific progress are. They are simply trying to accelerate production, which is like adding lanes to a highway when the slowdown is actually caused by a toll booth. It’s sure to make things worse. Contents 1. Science has been slowing — the production-progress paradox 2. Why is progress slowing? Can AI help? 3. Science is not ready for software, let alone AI 4. AI might prolong the reliance on flawed theories 5. Human understanding remains essential 6. Implications for the future of science 7. Final thoughts Full essay (about 6,500 words) aisnakeoil.com/p/could-ai-slo…

@panomics @GenomicsCow Was thinking of the CCDC genes too

@GenomicsCow Every one in human omics should be at >100 because they can state the olfactory receptor superfamily. Just ORIA through OR4X is there alone.

How many genes do you think you can name? Regardless of remembering its details

@Dr_ElvinaA @Larvol @OncoAlert @oncodaily Citation?

#ASCO25 may be over, but its impact continues. Colorectal cancer patients: 💍 Married: 5-year OS → 63.1% 🚶‍♂️ Single: 54.5% 💔 Separated/Divorced/Widowed: 45.7% ✅ Better support ✅ Earlier stage at diagnosis ✅ More treatment compliance @Larvol @OncoAlert @oncodaily