Principal AIDD 💊🤖Scientist at @InSilicoMeds 🧬, Retrosynthesis🧪CADD expertise TL // neo-hegelian, e/acc // All posts reflect my personal views only
Abu Dhabi, UAE
Joined April 2018
- Tweets147
- Following137
- Followers331
- Likes854
And it also getting cheaper! Still incomparable to light-weighted tools, but.. Results are soon available at dddbench.insilico.com, but here is the spoiler. I feel like camera-ready version of our NeurIPS paper on URSA bench will face many plot twists as new models emerge. 2/2
Plot twist level: “Diety”
In this era of artificial intelligence, it is becoming urgent to distinguish human art from what machines produce. There is an ontological difference, even before an aesthetic one, between art and what a machine can generate through statistical calculation based on millions of images created by others. Algorithms lack the spark of humanity. For this reason, the Church wishes to renew an alliance with artists and cultural institutions to safeguard our humanity.
The problem is that biology and chemistry are not as easily verifiable as math and coding. Formal logic is blind to uncertainties of nature. Robotics? You can never be 100% sure, if automated lab does makes an experiment that matches reality. Lab experiment != reality.
It is happening! GPT-6 Sol ☀️ matches the performance of the 🥇 SOTA synthesis-planning model RetroChimera, spotlighted in Nature just 4 days ago, in our URSA bench accepted for #neurips2026. The future of AI-retrosynthesis is fantastic! 1/3 🧵
Thrilled to share that our URSA benchmark #URSABench for multi-step retrosynthesis 🧪assessment has been accepted for #NeurIPS2026! Congrats 👏🥳 to the team @chem_logos @chemcensor @anmorgunov @Max__Kuznetsov @RimShayakhmetov @VladAladin @AlexAliper @biogerontology ! 🧵1/3
Here are the resources related to our submission, but there more at the camera-ready step:
📜URSA preprint: arxiv.org/pdf/2607.04688
🧪URSA Benchmarking sets: huggingface.co/datasets/insi…
🤖Our DDDBenchmark site for the most-up-to-date models' results: dddbench.insilico.com/
📊URSA GitHub Repo: github.com/insilicomedicine/… 🧵 2/3
Btw, #URSABench independently ranked RetroChimera by @MaziarzKris, @marwinsegler, and colleagues — recently published in Nature (nature.com/articles/s41586-0…) — 🥇 first among a range of open-source tools and frontier LLMs.
What still intrigues me is whether conventional tools can keep pace with rapidly advancing LLMs... Spoiler: GPT-6 Sol ☀️ is VERY good at synthesis planning, but in our benchmark it is extremely expensive compared with RetroChimera — not Pareto-optimal at all.
Still, I have a gut feeling that many things may change before we meet in Sydney 🇦🇺 in December... #NeurIPS 🧵 3/3
Huge congrats to @MaziarzKris @marwinsegler and the team with their RetroChimera published in Nature!
Btw, this model is still the most powerful open-source tool to perform synthesis planning for both OOD and in-distribution targets molecules according to our #URSAbench
Synthesis is a major bottleneck in drug discovery. AI synthesis planning can help propose better routes and make generative molecular design tractable.
With our retroChimera model we have been pushing the boundaries of these models further - out in @Nature today. 1/2
Bogdan A. Zagribelnyy, PhD retweeted
1.5 years ago, we bet the company on benchmarking and evaluating all things AI for science.
Today, we’re #1 in many areas. One of them is AI for longevity. Just ask your favorite LLM.
Here’s what GPT Astra says about our Cell cover paper 👇
Link in the comments
Nick Land once said that “the most robust definition of intelligence is competence at playing games.”
I hope I live to see the day when artificial intelligence becomes strong enough to win the Chess ♟️ of Aging ☠️ from the right side of the board.
Can you checkmate Aging before time beats you?
@elonmusk and @chesscom debated whether computers could solve chess.
10 minutes. Research tools for pieces. Biology makes the counter-moves.
Your move: agemate.org 👇
Extremely proud of my colleagues from @InSilicoMeds and collaborators from @liquidai, who delivered it. “AI against Aging” is the best anti-doomer vaccine. Kudos 🥳 to @biogerontology @Theo_Lotrus @vovalive @VladAladin @qoffee1337 @AlexAliper @ramin_m_h @xanamini @RimShayakhmetov @zulfatmif @mathieu_reymond
Aging has a language. 🧬 We’re teaching AI to read it.
On the cover of Cell: compact Longevity-LLMs, jointly trained by Insilico Medicine and @liquidai, demonstrate SOTA performance on LongevityBench.
Powered by these models, Longevity Claw nominated 328 candidate gene targets for aging research.
Small model. Big questions. Open source. 🧵👇
#insilicoSOTAFM
Actually adding single CH3, OH, Cl, F or Br is not randomness, especially for Me and Cl. “Magic methyl” and “magic chloro” are still there and the related affinity improvements are quite expected. Authors took “best practice” techniques and called it randomness… wierd
Before claiming your AI understands chemistry, prove it does better than randomness.
nature.com/articles/s41586-0…
“Drugs are the new oil” — and the UAE 🇦🇪 is doing fantastic things to prove it.
Mendeleev once famously said that “burning oil is like burning banknotes.” When oil is finally transformed into medicines, both its economic value and its value to humanity grow exponentially.
🇦🇪 New AI drug discovery milestone from the UAE: BTK-targeting PROTACs for potential applications in blood cancers.
Our UAE team used AI to help design a new series of BTK degraders.
🧬 Lead compounds showed picomolar BTK degradation
💊 Oral bioavailability in mice
🎯 Activity against wild-type and mutated BTK
🤖 Novel building blocks designed with Chemistry42
The research was published in the Journal of Medicinal Chemistry and builds on Insilico Medicine’s growing drug discovery efforts in the UAE.
From designing molecules locally to publishing internationally, this is another step toward building AI-powered drug discovery capabilities in Abu Dhabi. 🇦🇪
Bogdan A. Zagribelnyy, PhD retweeted
The LLM race is getting interesting in a different way: not just smarter models, but better tools.
GPT-6 Astra keeps surprising me with what one prompt can produce.
Here's @InSilicoMeds' ChemCensor assessing reactions against real synthetic precedents.
@OpenAI @joyjiao12
The history behind Rentosertib is mostly about multi-parameter optimization. Finally, after the MPO the best compound should be a summit 🏔️ on the 3D landscape collected layer by layer. Something that cannot be caught just from looking at excel tables.
52 compounds. 20 property maps. One landscape.
A scientific tool built around one very specific question I had.
Here’s @InsilicoMeds’ rentosertib series, visualized with GPT-6 Astra. 👇