Bioinformatician, Cellular Biologist, CSO @tychobio_ai @mathurindorel@fediscience.org @mathsrish.bsky.social
Berlin, Germany
- Tweets 13,435
- Following 1,657
- Followers 418
- Likes 82,082
Pinned Tweet
Midnight reflexion on to improve scientific evaluation beyond h-index: how about a veting mechanism with pagerank based on e.g ORCID profiles.
Each researcher can vouch for any number of other researchers. The importance of a researcher is then simply their pagerank.
1
1
3
2,317
I'll be in SF from Thursday 1st to Sunday 4th.
If any folks there want to grab a drink or has a good biofounder event suggestion, ping me.
1
1
2
234
Are you kidding me? They have a fricking correct statistics discussion about power and alpha!!
Pretty unrealistic given they're biologists but still whoever consulted on the science deserves a vulgarisator award.
How a fan fic about the starwars sequels end up being a great romantic comedy about the power balance in academia is fascinating evolution.
Quite impressed by the depth of some discussions in The Love Hypothesis.
Pretty good nerdsnippe for any MAPK fan out there as well.
1
105
How a fan fic about the starwars sequels end up being a great romantic comedy about the power balance in academia is fascinating evolution.
Quite impressed by the depth of some discussions in The Love Hypothesis.
Pretty good nerdsnippe for any MAPK fan out there as well.
1
156
Mathurin Dorel retweeted
It’s interesting that the language being used here removes agency from the labs
I frequently hear “agents are good at escaping sandboxes”
And not “labs test agents highly insecure environments to test agents in, resulting in huge externalities”
If Boeing missiles regularly misfired into major cities during the R&D process, no one would say “missiles are really good at accidentally misfiring”
The headline would definitely be “Boeing fails to safely develop missiles”
OpenAI and Anthropic are now investigating ***tens of thousands*** of incidents - not dozens.
"The total could grow well beyond tens of thousands."
"The incidents range in severity and are comparable to disclosures by OpenAI in recent days. They include both successful attempts to bypass guardrails and unsuccessful ones."
"The sheer number of incidents indicates that the problem is orders of magnitude more complex than what is publicly known."
27
116
793
34,314
Mathurin Dorel retweeted
I’m a scientist. I need to say this because the AI hype is getting ridiculous.
AI can design a molecule in seconds.
That doesn’t mean it discovered a drug.
It discovered something we scientists have never been short of:
Something to test.
Someone still has to make it.
Run the experiment.
Measure whether it works.
Check whether it’s toxic.
And ultimately prove it works in the real world.
AI hype tells us:
“Prediction is discovery.”
“Simulation is experimentation.”
“Generating a molecule is developing a drug.”
It isn’t. AI is making ideas incredibly cheap.
But every new idea creates something AI cannot generate: Evidence.
And the more hypotheses AI produces, the more experiments we’re going to need.
That’s the irony nobody seems to be talking about.
AI may not make laboratories obsolete.
It may make them more valuable than ever.
You can speedrun the thinking.
You can’t speedrun reality.
150
538
1,813
56,576
Mathurin Dorel retweeted
Despite all the recent backlash against ASOs (due to a few bad trial outcomes), they still show they are a great class with the success of Ionis's Ulefnersen for FUS-ALS when Novartis mAb VHB937 failed
1
2
3
1,116
I mean, definitely cool.
But also why would I use this over store.nanoporetech.com/us/di… for any of the applications mentioned ?
Today, Sculpta releases the first demonstration of bioorthogonal barcoding ("bobcoding"), the process by which we:
1) covalently attach oligonucleotide barcodes to RNA at multiple internal positions with two-step chemistry, and 2) perform a multiplexed reverse transcription reaction to generate barcoded cDNA libraries with >99% barcoding accuracy.
To the best of our knowledge, it is the first time ever that either of these steps have ever been performed by anyone. No one has ever broken the barrier of multiplexing prior to any enzymatic step in library preparation, driving simplicity and higher data quality. Instead of other methods that add 0 or 1 barcode, we add one barcode every ~300bp, adding redundancy and fidelity to RNA measurements.
In our first application described in our pre-print below, we used bobcodes to create the first RNA isoform-resolved drug screening platform. It beats Novartis' DRUG-seq platform by:
- generating full length transcript capture of RNA isoforms, instead of just 3' end counting
- capturing 25-fold more RNA splicing events
- reducing barcode swapping by 10-fold
- using 18 fewer PCR cycles (~250,000 less amplification)
- reducing sample-to-sample variability in gene expression measurements
- eliminating costly and cumbersome library fragmentation/tagmentation steps completely
- reducing workflow complexity and number of steps
- reducing overall protocol duration by ~25%
Though our chemical barcoding method improved transcriptomic data quality while also being simpler and faster than all existing methods, the implications of this work go well beyond this particular use case:
The gate to greater applications of AI in transcriptomics is not a lack of compute. Nor is it a lack of data volume. The elephant in the room has always been our limited ability to faithfully and accurately measure cellular RNAs.
Sculpta now has line of sight to build what we call the first ground truth transcriptomics platform [for single cell, spatial and more] for the future of biological research, drug development and AI-enabled discoveries for the betterment of human health and longevity.
[PS Until the kind folks at bioRxiv get through the apparent backlog of AI slop submissions, Sculpta will host the pre-print PDF on our website. You can sign up for product offerings and other updates at this link too]
sculpta.bio/preprint
1
282
@AnthropicAI @OpenAI
Get better or raise less, your call.
We do hope you get better, the tools you created are tremendously useful and should be leveraged ethically.
1
19
Certes la citation est marrante mais ce serait bien que les gens comprennent un peu l'entraînement des modèles.
Les versions de Claude depuis Mythos sont entraînés sur des données de séquences génétiques en plus du texte humain.
Tout comme openAI a initié un focus sur le lean.
2
291
Autonomous, but not trusted?
Room to scale up too, are some devices recurrently rate limiting or is everything optimized?
In case you are wondering what the schedule / gantt chart looks like for 70+ different lab protocols running on an autonomous lab right now. Each row is a device, different colors are different protocols submitted by scientists at ginkgo today. It's getting very cool 😍
1
1
1,859
Interesting result by Anthropic.
Beyond the "new phage array based nuclease" hypothesis, it is an interesting piece on harness refinement for biology (we're not there yet), model capability to deeply understand genomic data (just breached) and the role of specialists (still)
Claude has discovered a previously unknown enzyme system hidden in the DNA of bacteriophages. Beside the enzyme’s gene sits a long array of repeating DNA—a structure that looks somewhat similar to CRISPR.
We don’t yet understand what this system does, but only a handful of known systems share its features, and all of them are able to cut, copy, and paste DNA. Historically, the discovery of such programmable systems has helped revolutionize medicine. CRISPR, for instance, is now the foundation of genetic medicines. But it will take much more work to learn what this system does, and whether it can be put to similar use.
Read more: anthropic.com/news/claude-di…
1
2
562
Mathurin Dorel retweeted
The RNA space is at a crossroads.
We analyzed >3.1K therapeutic oligonucleotide programs for Sleuth's RNA report. The headline is that within the liver, RNA is moving towards a product execution story, but beyond it, it's still a delivery one.
Here's the story in a nutshell:
1. Industry pioneers like Alnylam and Ionis had to first prove synthetic oligos could reliably silence a target. Alnylam's first approved RNAi drug Onpattro used lipid nanoparticles to deliver siRNA.
GalNAc made liver delivery repeatable, turning RNA into a legitimate platform that could hit a variety of liver-expressed targets, which we saw Alnylam and Ionis take advantage of for a combined 8 approvals between 2019 - 2025.
2. But with liver delivery solved via widespread GalNAc adoption (+ standardized sequence design & chemical modification), many drug developers ended up converging on the same targets. And GalNAc conjugate patents are starting to expire in 2028, so the competitive pressure will only accelerate.
With some programs already showing 90+% knockdown and 6mo durability, incremental knockdown is just less interesting, pushing genuine differentiation towards traditional clinical metrics.
3. In response, developers have generally arrived at two potential paths:
(a) 𝗶𝗻𝗱𝗶𝗰𝗮𝘁𝗶𝗼𝗻 𝗱𝗿𝗶𝘃𝗲𝗻 𝘀𝘁𝗿𝗮𝘁𝗲𝗴𝗶𝗲𝘀: use proven liver delivery to pursue larger chronic indications and compete with the broader market / established SoC on clinical outcomes, safety, dosing, patient burden, cost and overall value.
(b) 𝘂𝗻𝗹𝗼𝗰𝗸 𝘁𝗵𝗲 𝗻𝗲𝘅𝘁 𝘁𝗶𝘀𝘀𝘂𝗲: make the bio-engineering leaps needed to allow systemic delivery into muscle, kidney, CNS or other tissues as repeatable as GalNAc into the liver.
There's certainly a lot of upside with path two, but there's a long list of challenges to address:
• cell-specific uptake
• productive intracellular delivery
• endosomal escape
• adequate activity, durability and safety
• scalable manufacturing
As shown in the report, only ~12% of clinical programs are clearly attempting to do this, although that figure is much higher for early preclinical programs.
There's clearly demand from Pharma to pay "platform prices" when a capability works across assets - Novartis' $12B acquisition of Avidity is one example, but you need to separate the clinical outcome of a single asset (del-desiran Ph3 HARBOR study missed its primary functional endpoint) from whether muscle-directed AOCs work. A big part of the payment is to buy a way into this new tissue.
The key for the field is to reproduce the success the last generation of programs showed: a genuine platform where the 2nd and 3rd drugs use the same route into a tissue and deliver clinical benefit.
The full report touches on deep dives into CNS / muscle delivery, 2026-2027 catalyst calendar and RNA dealmaking since 2021, among other things.
If you'd like it, just comment "RNA" below and we'll send it over!
75
8
76
10,714
Mathurin Dorel retweeted
The takes on Terry Tao being unemployed seem to me to be an expression of mediocrity and resentment, from people who have never tasted dedication to something higher than themselves. The guy is not going to lose his job. He is forever in the Hall of Fame. If he was purely selfish, he'd be grinning that he got on the train right before it crumbled.
Maybe, just maybe, he's sad to see something he cared about potentially crumble, instead of being self- interestedly afraid "for his job".
49
129
1,685
58,784
It looks like Opus 5.5 time to shine, with Luna not far behind.
Astra held less than a month.
I don't think slowing down is on the menu.
60
It's a beautiful paper of codon optimisation for an apparently simple task: maximise the number of paired bases in an mRNA. Solved with elegant semantic graph mathematics.
Choose the ratio between thermal stability and translation efficiency for a vaccine
nature.com/articles/s41586-0…
Suppose we could keep mRNA vaccines effective without expensive refrigerated storage and transport. We could make vaccines against flu and Covid, and potentially future personalized cancer vaccines, easier to deliver to more people. Fewer doses would end up spoiled and thrown away.
Well, a new preprint from MIT takes a step toward that. The researchers made dried mRNA Covid vaccine formulations that, after two months at 37°C, still produced antibody responses in mice comparable to those produced by fresh vaccine.
AI helped find the right recipe. Drying a vaccine without damaging it requires the right protective ingredients in the right amounts. The possible combinations quickly become too numerous to test individually. The team used machine learning to learn from each round of experiments and choose which mixtures to test next.
And if that wasn’t enough, the same paper also reports delivering the stabilized vaccines through patches covered with microscopic needles that dissolve in the skin. In mice and monkeys, these produced immune responses similar to those produced by injected vaccines.
The connection is practical: a vaccine that stays stable in a dry patch could be easier both to transport and to administer, potentially reducing the need for trained staff to give injections. (3/7)
1
695
The group better be well paid to check the proof.
And I wouldn't be surprised if they find that some are wrong and start with Ctrl-F "sorry".
We’re working with an independent advisory group of mathematicians to help OpenAI responsibly share advances in AI and mathematics.
The group will advise on how we assess and communicate new mathematical results, uphold academic and professional standards, and build tools that support mathematical research and learning.
Through this work, we want mathematicians to be at the center of shaping how AI supports mathematical understanding and how its benefits reach the wider community.
openai.com/index/advisory-gr…
3
2
701
