Lightnews — Scholar-powered news

Marwin Segler

@marwinsegler.bsky.social

4.2K followers 880 following 180 posts

Machine Learning, {Org, Med, Comp} Chem, RL/Planning, AI-assisted Scientific Discovery & Creativity, Music. ELLIS Scholar. Team Lead at Microsoft Research AI for Science. 2xDad

Posts Media Videos Starter Packs

Pinned

Marwin Segler @marwinsegler.bsky.social · Dec 9

new preprint on chemical synthesis ML models

- showing how to combine multiple models in a principled way
- modern Transformers + GNN to featurize chemical reaction:
- new insights in where the models shine
+ bonus: find the quirky named reaction!

Feedback welcome!

arxiv.org/abs/2412.05269

Chimera: Accurate retrosynthesis prediction by ensembling models with diverse inductive biases

Planning and conducting chemical syntheses remains a major bottleneck in the discovery of functional small molecules, and prevents fully leveraging generative AI for molecular inverse design. While ea...

arxiv.org

4 27 83

Marwin Segler @marwinsegler.bsky.social · 8d

Have you also considered the cost of classical or quantum-chemical simulations? Those require supercomputers as well

3 2

Marwin Segler @marwinsegler.bsky.social · 29d

What’s you take?

Marwin Segler @marwinsegler.bsky.social · Sep 4

And for next year, Skala?

1 1

Marwin Segler @marwinsegler.bsky.social · Aug 5

Are they also playing against the latest version of Stockfish with similar compute budget?

Reposted by Marwin Segler

Tian Xie @xie-tian.bsky.social · Jul 31

Want to join our efforts @msftresearch.bsky.social AI for Science to push the frontier of AI for materials? We are the team behind MatterGen & MatterSim and we have 2 job openings! Each can be in Amsterdam, NL, Berlin, DE, or Cambridge, UK.

Search Jobs | Microsoft Careers

jobs.careers.microsoft.com

1 5 7

Reposted by Marwin Segler

Nancy Baym @nancybaym.bsky.social · Jul 28

We may have the chance to hire an outstanding researcher 3+ years post PhD to join Tarleton Gillespie, Mary Gray and me in Cambridge MA bringing critical sociotechnical perspectives to bear on new technologies.

jobs.careers.microsoft.com/global/en/jo...

Search Jobs | Microsoft Careers

https://jobs.careers.microsoft.com/global/en/job/1849026/Principal-Researcher-–-Sociotechnical-Systems-–-Microsoft-Research

49 89

Reposted by Marwin Segler

Jake Yeston @jakeyeston.bsky.social · Jul 27

I put this up at the old place in 2019 when we did our Periodic Table anniversary special issue, with my own added lyrics for all the post-Nobelium elements

chemsky

1 3 15

Marwin Segler @marwinsegler.bsky.social · Jul 26

The whole channel is great 😂

Marwin Segler @marwinsegler.bsky.social · Jul 23

awesome initiative!

Marwin Segler @marwinsegler.bsky.social · Jul 22

Best of both worlds then 🙂

Marwin Segler @marwinsegler.bsky.social · Jul 22

Welcome back to 🇬🇧🙂

1 1

Marwin Segler @marwinsegler.bsky.social · Jul 21

Have the students all moved to voice-controlled vibe coding, without need for keyboards?

1 4

Marwin Segler @marwinsegler.bsky.social · Jul 19

Which one are you reading?

Marwin Segler @marwinsegler.bsky.social · Jul 19

Can almost taste it from the photos, fantastic!

Marwin Segler @marwinsegler.bsky.social · Jul 18

One of my favorite regions, enjoy! 🍇

1 1

Reposted by Marwin Segler

Stephanie Hyland @hylandsl.bsky.social · Jul 18

New work from my team! arxiv.org/abs/2507.12950
Intersecting mechanistic interpretability and health AI 😎

We trained and interpreted sparse autoencoders on MAIRA-2, our radiology MLLM. We found a range of human-interpretable radiology reporting concepts, but also many uninterpretable SAE features.

Insights into a radiology-specialised multimodal large language model with sparse autoencoders

Interpretability can improve the safety, transparency and trust of AI models, which is especially important in healthcare applications where decisions often carry significant consequences. Mechanistic...

arxiv.org

1 4 11

Reposted by Marwin Segler

Peter Kenny @pwk2024.bsky.social · Jul 18

Özçelik et al (2025) Generative Deep Learning for de Novo Drug Design─A Chemical Space Odyssey. JCIM ASAP #cheminformatics #CompChem #ML #AI #OpenAccess #chemsky 🧪
doi.org/10.1021/acs....

Generative Deep Learning for de Novo Drug Design─A Chemical Space Odyssey

In recent years, generative deep learning has emerged as a transformative approach in drug design, promising to explore the vast chemical space and generate novel molecules with desired biological pro...

doi.org

4 12

Marwin Segler @marwinsegler.bsky.social · Jul 12

Love the blog post (JPO as a random English guy 😂), Hope you have a great time at the conference!

Marwin Segler @marwinsegler.bsky.social · Jul 11

Ich hoffe dass alle Kandidaten eine Einstellungsgesprächsdiskussionsgrundlagenpublikationsvortagspitzenleistung abgelegt haben!

Marwin Segler @marwinsegler.bsky.social · Jul 10

Hat aber mehr Material als der preprint! @franknoe.bsky.social

1 2

Marwin Segler @marwinsegler.bsky.social · Jul 10

Scalable emulation of protein equilibrium ensembles with generative deep learning

www.science.org/doi/10.1126/...

Scalable emulation of protein equilibrium ensembles with generative deep learning

Following the sequence and structure revolutions, predicting functionally relevant protein structure changes at scale remains an outstanding challenge. We introduce BioEmu, a deep learning system that...

www.science.org

1 3 14

Reposted by Marwin Segler

Adam Foster @adamefoster.bsky.social · Jun 26

I am very happy to share Orbformer, a foundation model for wavefunctions using deep QMC that offers a route to tackle strongly correlated quantum states!
arxiv.org/abs/2506.19960 (1/n)

An ab initio foundation model of wavefunctions that accurately describes chemical bond breaking

Reliable description of bond breaking remains a major challenge for quantum chemistry due to the multireferential character of the electronic structure in dissociating species. Multireferential method...

arxiv.org

1 8 29

Marwin Segler @marwinsegler.bsky.social · Jun 22

a lot of this functionality is also in the guacamol package, see also the discussion the guacamol paper + SI data for molecule "sanity" checks

1 1

Marwin Segler @marwinsegler.bsky.social · Jun 18

One for #compchemsky

Jan Hermann @jan.hermann.name · Jun 18

🚀 After two+ years of intense research, we’re thrilled to introduce Skala — a scalable deep learning density functional that hits chemical accuracy on atomization energies and matches hybrid-level accuracy on main group chemistry — all at the cost of semi-local DFT ⚛️🔥🧪🧬

1 10

Reposted by Marwin Segler

Jan Hermann @jan.hermann.name · Jun 18

3 25 72