🔗 Webpage: aimat.iar.kit.edu/ml4mol2026.php

I believe the school is open to all, but especially relevant for PhD students in their first year(s) and Master's students who are interested in doing a PhD in one of the areas covered by the school!

Sign up before June for full consideration.

Registration is now open for the AiMat Summer School "Machine Learning for Molecules" 2026, organized in Karlsruhe 14-18 September by Pascal Friederich & team. I will be one of the lecturers. Sign up, and/or share with your students, for what is sure to be an exciting week of tutorials! 💻 #chemsky

We benchmarked 4 LLMs vs 17 expert chemists on retrosynthetic route evaluation.

tl;dr—Gemini 2.5 Pro leads, GPT-o3 is too pessimistic, Claude & GPT-4.1 too optimistic. Interestingly, the more balanced models change their score when asked over different instances.

Read on for more details!

🧪 Can LLMs judge chemical synthesis plans?

Excited to share new work from our industrial postdoc Varvara Voinarovska, in collaboration with Samuel Genheden and Mikhail Kabeshov at AstraZeneca's Molecular AI team!

📄 : doi.org/10.26434/che...

Thanks for the excellent talk today @briantrippe.bsky.social ! We will post the Zoom recording soon

Glad to share our new work: RitS (Right into the Saddle): a generative model for direct transition state (TS) prediction.
📄 Paper: chemrxiv.org/doi/full/10.26…
🤠 Code: github.com/isayevlab/RitS
• Flow-matching model with explicit stereochemistry control
• Direct TS generation from SMILEs #compchem

We are looking forward to welcome @briantrippe.bsky.social for this months Chalmers AI4Science seminar tomorrow. Check psolsson.github.io/AI4ScienceSe... for details on how to connect!

What if AI could invent enzymes that nature hasn’t seen? 👩‍🔬🧑‍🔬

Introducing 🪩 DISCO: Diffusion for Sequence-structure CO-design

📝 Blog: disco-design.github.io
📄 Paper: arxiv.org/abs/2604.05181
💻 Code: github.com/DISCO-design...

Adaptive, Bayesian Experimental Design to Efficiently Determine the Critical Micelle Concentration of a Surfactant Surfactants are widely used for industrial applications, yet more environmentally friendly surfactants with enhanced properties are demanded. A key thermodynamic property governing the behavior of a surfactant in an aqueous solution is its critical micelle concentration (CMC). Below the CMC, increasing the surfactant concentration reduces the surface tension of the solution; above the CMC, the water–air interface becomes saturated with adsorbed surfactant, leading excess surfactant to self-assemble into micelles and the surface tension to plateau. Many physicochemical properties of a surfactant solution exhibit sharp changes at the CMC. The conventional experimental protocol to determine the CMC of a surfactant is labor-intensive and time-consuming: (1) prepare many surfactant solutions spanning a wide concentration range and then (2) measure the surface tension of each solution. Herein, we adopt Bayesian experimental design (BED) to determine the CMC of a surfactant more efficiently─even without prior knowledge of its order of magnitude. BED follows an experiment-model-design feedback loop: (1) prepare a surfactant solution and measure its surface tension; (2) use all surface tension data thus far to obtain a posterior distribution over thermodynamic models of the surface tension isotherm of the surfactant; and (3) pick the surfactant concentration for the next experiment to maximize expected information gain about the CMC. We show that BED efficiently gathers information about the CMC using two surfactants (octyl-β-d-thioglucopyranoside and Triton X-100) as test cases. Broadly, BED can reduce the time, effort, cost, and chemical waste to determine the CMC of surfactants and drive an autonomous laboratory for surfactant discovery and characterization.

check out our new paper on Bayesian experimental design for surfactant characterization!

pubs.acs.org/doi/10.1021/...

We validated via rediscovery of fluorinated/phosphorus-containing solvents, then proposed novel candidates for fluorinated diluents and non-fluorinated weakly solvating electrolytes.

Work initiated through a WASP/WISE pilot project (now continuing our collaboration as part of a NEST project) 🙏

Key ingredients: a curated battery-specific dataset (115K molecules), physics-informed ML surrogates for redox potential, viscosity, melting point, donor number, and dielectric constant—all plugged into an RL loop to steer generation toward viable candidates. ⚛️

New preprint with Chao Zhang's team at Uppsala University! We used GraphINVENT to design new rechargeable battery solvents de novo. 🧑‍💻 #chemsky

Pre-print: doi.org/10.26434/che...
Code and data: to be released

Join us in Gothenburg! Excellent conditions, scientific independence, and low teaching load on the tenure track. Feel free to reach out if you have any questions.

My main gripe with the alphafold example is how it shows you need decades and decades of high quality data, well structured, open and accessible to train a model -- and yet they always gloss over it and pretend it's just AI and magic. No, we need to continuously invest in real data and FAIR data.

Careers – aichemy Join the AIchemy team and become part of a world-class research and innovation community. We are building a dynamic group of innovators, researchers and professionals committed to advancing the…

🚨 PhD Opportunities at the University of Liverpool 🚨

There are 4 PhD positions available at the cutting edge of AI, robotics, and chemical discovery 🧪🤖

👉 For more information

Recent advances in modeling and simulation of biological phenomena in crowded and cellular environments While experiments and computer simulations to study biological phenomena are usually performed in diluted in vitro conditions, such phenomena happen inside the cellular cytoplasm, an environment dense...

Happy to share the review that @apoorvam.bsky.social, Vanessa and I wrote about recent trends in modeling and simulation of biological phenomena in crowded and cell-like environments! I really like this topic, for me this is the future of #compchem.
arxiv.org/abs/2603.26974

Banner of speakers for the ML2MD symposium.

Registration is NOW OPEN for the 2026 Machine Learning for Materials and Molecular Discoveries (ML2MD) Symposium in Uppsala 🇸🇪 to be June 17-19 (right before midsummer 🌼), which I'm co-organizing w/ Chao Zhang! Come join us for 3 exciting days exploring the interface of ML and materials! #ML2MD2026

15th RDKit UGM 2026 2026 RDKit User Group Meeting (in person and online)

Free registration for the 2026 #RDKit UGM (both in-person and online attendance) is now open:
www.eventbrite.com/e/1985889262...

Registration is open for the Strasbourg Summer School in Chemoinformatics 2026. 👉 infochim.chimie.unistra.fr/-strasbourg-...

We had our 2026 team workshop yesterday for my lab + co-supervised students. Lots of team building and tutorial activities followed by dinner. Really proud of the team we’ve built working on AI for molecules, and so grateful for all the support we’ve received along the way!

Sander Dieleman - Diffusion models for image and video generation | ML in PL 2025 YouTube video by ML in PL

In October, I gave a talk at ML in PL in Warsaw: a whirlwind tour of what goes into training image and video generation models at scale.

📺 video: www.youtube.com/watch?v=qFIT...
🖼️ slides: docs.google.com/presentation...

2026 Machine Learning for Materials and Molecular Discoveries Symposium About the Event Join us for the 2026 symposium on Machine Learning for Materials and Molecular Discoveries (ML2MD). At this symposium, we seek to highlight some of the key research developments at the...

👉👉👉 For more details and registration, please see link below:
chalmers.ungapped.io/Events/6bdec...

Hope to see many of you in Uppsala for this event, maybe you stick around for midsummer celebrations in Sweden after 🌼🌸🇸🇪

Please share and help me get the word out about this event! #chemsky #chemchat

Very grateful to all our sponsors who are supporting this event, including the Wenner-Gren Foundations, the WASP & WISE programs, and the DAEMON EU Cost Action. 🙏

More info on each sponsor can be found on our event page: chalmers.ungapped.io/Events/6bdec...

(speakers continued)
@kjablonka.com
Nicholas Jackson
Maxime van der Heijden
Matti Hellström
& Harish Gudla

Really looking forward to meeting all in person in Sweden, in the best time of the year, and hearing all about your science! 🗣️🔬🧑‍💻

Full speaker list:
chalmers.ungapped.io/Events/6bdec...

Amongst the amazing line-up of speakers we have:
Rampi Ramprasad
@jelfschem.bsky.social
Cheng Shang
Moyses Araujo
Anders Hellman
Xiang Chen
@juliawiktor.bsky.social
Ying Li
@shijingsun.bsky.social
@aduvalinho.bsky.social
@leitingzhang.bsky.social
(continued below)

Banner of speakers for the ML2MD symposium.

Registration is NOW OPEN for the 2026 Machine Learning for Materials and Molecular Discoveries (ML2MD) Symposium in Uppsala 🇸🇪 to be June 17-19 (right before midsummer 🌼), which I'm co-organizing w/ Chao Zhang! Come join us for 3 exciting days exploring the interface of ML and materials! #ML2MD2026

Katharina with her award certificate in front of her poster

A poster award for Katharina Ueltzen at the Chemical Compound Space Conference! So very happy for her and I think it is so very much deserved. She drives the project and I am very lucky to work with her.

congrats! 🎉

And some pictures from today @jtmargraf.bsky.social @joshuaschrier.bsky.social @grynova.bsky.social thanks for a great meeting especially to the organizers 🙏 #CCSC2026

Some more highlights from the afternoon session yesterday @podewitzlab.bsky.social @thijsstuyver.bsky.social and many more great talks 🤩