Eternal skeptic, occasional optimist. AI, Bio, Chem, and side quests.
Montréal, Québec
Joined January 2015
- Tweets238
- Following389
- Followers457
- Likes714
Pinned Tweet
Ever heard of Spasfon, the pink pill? It has been prescribed for more than 60 years, especially to women. The catch? Clinical evidence for many of its common uses is remarkably weak.
Ep 2 of Beyond the Assay is out: enoutahi.com/blog/spasfon-fa…
This weekend, I used frontier LLMs to screen the math claims in ~4,800 accepted NeurIPS 2026 papers, over half across all tracks.
Agents read the claims, and when something doesn’t hold, they explain why, check whether it was already addressed elsewhere, and propose a repair when they can. A second set of agents then checks the criticism and refine it.
14% of the assessed papers were flagged for major or critical issues.
proofnote.enoutahi.com/
3
9
211
This is less exciting than solving Millennium problems, but speeding up science a little still seems fun.
Maybe @NeurIPSConf could run automated checks on mathematical claims during review ? And frontier labs could sponsor it 😁 ?
1
38
If you (or your AI agents 😉) use the Datamol ecosystem, I updated datamol, molfeat, medchem and SAFE a few weeks ago: bug fixes, new featurizers, and support for latest Python and RDKit versions.
SAFE also fixes some edge cases where SMILES ↔ SAFE conversion lost tetrahedral or E/Z stereo.
github.com/datamol-io/
"The Machine Stops" by Forster is one of those stories that feels more contemporary every time you return to it cs.ucdavis.edu/~koehl/Teachi…
Written in 1909, but it feels so current....
1
88
I am throwing random tasks at GPT-6 #Astra, including #chess
Too slow for blitz against humans (through computer use) even when told to play faster.
In King of the Hill on @lichess, it lost to Stockfish level 7. I then lowered the difficulty to level 3, switched from default to X-high reasoning, and explicitly told it to do better and take its time thinking.
It won nicely.
3
176
1/ Looking back at drug marketing from the 1960s and 70s is… an experience.
Here a Thorazine ad literally placed African masks and ritual objects under the heading “basic tools of primitive psychiatry,” in contrast with Thorazine, presented as a tool of Western medicine.
2/ I’m starting a blog series called Beyond the Assay, about drugs that changed medicine in ways few could have anticipated, and what their stories might teach us about drug discovery in the age of AI.
The first essay is about Thorazine / chlorpromazine:
enoutahi.com/blog/largactil-…
1
1
161
Incredible work from my colleagues @NikhilShenoy12 and @Francesco_dgv ! And this is only a glimpse of what they have been cooking lately🔥
To identify the most promising drug candidates, we need to understand not only whether a molecule is likely to bind to a target, but how strongly.
Today, we’re excited to introduce Nesso-1, an open-source, coarse-grained co-folding model that significantly accelerates binding-affinity predictions.
7
558
Shocking new result 🙃: telling a deep learning model what we already know about gene interactions helps it predict biology it hasn't seen before. Until high quality biological data catches up, scaling might not be the only path forward.
1/ Our most recent Inside Valence blog post delves into TxPert, a SOTA model published in @NatureBiotech.
The key finding: scale alone is insufficient, even smaller model architectures can achieve top performance when coupled with biological priors.
@IhabBendid35780
Connecting molecular mechanisms to phenotypic outcomes is the missing piece for truly useful virtual cells. This work led by @yunhuijang_ shows we can structure LLM bioreasoning into verifiable, grounded mechanistic explanations that improve perturbation response prediction.
To realize our vision of better and more effective drug discovery, we must turn predictions into testable hypotheses and experimental designs.
Dive into the Explain pillar of our Virtual Cell framework: why it matters and how we adapt LLM reasoning to biology.
open.substack.com/pub/valenc…
Learn more here:
- Blog post : open.substack.com/pub/valenc…
- Paper: arxiv.org/abs/2604.11661
1
18
Nice overview as usual, @nathanbenaich. It left me wondering where durable value accrues today: models, infrastructure, or distribution?
One step closer to replacing the 'Labs' in our name with 'Simulations'
Even with today's industrialized labs that can run millions of experiments each week, we can only cover a tiny fraction of all the experiments we may want to run.
In this week's Inside Valence, learn more about virtual assays.
open.substack.com/pub/valenc…
1
67
Virtual cells shouldn’t just be predictive models,
they should be testable theories of cell biology.
They should predict cellular responses, explain them mechanistically, and discover new biology and therapies.
Excited to share our perspective paper: arxiv.org/pdf/2505.14613
1/ At Valence Labs, @RecursionPharma's AI research engine, we’re focused on advancing drug discovery outcomes through cutting-edge computational methods
Today, we're excited to share our vision for building virtual cells, guided by the predict-explain-discover framework 🧵
🚀 Polaris is here! It's time to stop overfitting on the same 3 benchmarks in drug discovery.
1/ 🚀Today, we’re excited to launch Polaris! 🚀
Polaris is a platform where the ML community working on drug discovery problems can easily share and access datasets and benchmarks.
Get started with just a few lines of code: polarishub.io/
This is just the beginning. Imagine the leaps ahead with LOWE as we transition from an LLM to a multimodal foundation model of biology and chemistry.
The cat is out of the bag !
The future of drug discovery is here.
LOWE (LLM-Orchestration Workflow Engine) is a cutting-edge software designed to perform complex #drugdiscovery tasks using a natural language interface.
🎥 Discover what LOWE can do ⬇️
Learn more: bit.ly/41SA9Pi
#JPM24 #TechBio
1
10
627
One of the nice things about Medchem is that it's designed for all stages of drug discovery, with a very intuitive API. Please check out the tutorials! #cheminfo
Introducing Medchem, an open source Python library for molecular filtering and prioritization.
It includes hundreds of molecular filters, alerts, and rules, both established and novel.
Read the blog by @HadiM_ for more details: portal.valencelabs.com/blogs…
🔥🔥🔥 this is part of what makes doing research at Valence Labs so unique.
The expansion of our supercomputer with @nvidia will add more than 500 NVIDIA H100 GPUs to the 300+ NVIDIA A100 GPUs that are already in place!
Stay tuned to learn more about the foundation models that we are building for drug discovery on top of industry-leading datasets. 👀