Nishi

Visiting Researcher @OxfordChemistry | Marie Curie PhD Fellow @univgroningen | Computational chemist | Previously @AganithaAI @uutah

[Nishi @nishimaru30]

guys posting "we are so back" - you know self-driving labs already exist, right? :)

[Anthropic @AnthropicAI]

Today, we're kicking off the first phase of the research preview for Model Hardware Standard (MHS): a new standard for AI agents to safely operate physical equipment in scientific research and advanced manufacturing.

Read more: https://t.co/XQ2y9EW7Af

[Kerala Tourism @KeralaTourism]

We made chips and it was crispy!

[Nishi @nishimaru30]

@b1shtream @cogworksai @furst_fly Good luck! 😄

[Prof. Anima Anandkumar @AnimaAnandkumar]

Tackling a 60-year-old challenge in quantum chemistry: making density functional theory scale nearly linearly with system size.

This has huge implications for opening the door to realistic systems that have traditionally been too expensive to simulate.

AI has attempted to accelerate these calculations, but models generally struggle to extrapolate, particularly to systems larger than those seen during training. Unlike text or images, quantum-mechanical training data is extremely expensive to generate.

We present a single unified AI model that performs quantum-mechanical simulations of both molecules and materials in quasi-linear time.

Using a novel Fourier neural operator variant, we learn the underlying Kohn–Sham equation map to produce physics-informed, self-consistent answers.

Unlike prior AI approaches that directly predict chemical properties, our model works through intermediate steps to improve difficult predictions, resembling inference-time reasoning in large language models.

To demonstrate its scalability, we run a self-consistent calculation of a magnesium dislocation with about 80k electrons on a single GPU, which previously required about 7k GPUs.

https://t.co/5DG12E04Rh

@DanishK42 @Caltech

[Sergio Ferrero @calotonterias]

Las personas que hablan 2 idiomas tienden a tener cerebros 6 años más jóvenes de lo que correspondería a su edad; las que hablan 3 idiomas, cerebros 7 años más jóvenes; y las que hablan 4 idiomas, cerebros 13 años más jóvenes.

El efecto protector del multilingüismo parece ser dosis-dependiente: cuantos más idiomas se hablan, mayores serían los beneficios. Este efecto se ha relacionado con varias vías, como una mayor estimulación cognitiva general, el aumento de la plasticidad cerebral al aprender estructuras lingüísticas diferentes, y factores sociales y culturales asociados al uso de varios idiomas.

Además, una de las vías principales sería que, cuando una persona maneja varias lenguas, todas permanecen activas en el cerebro al mismo tiempo, aunque en ese momento solo esté usando una. El cerebro tiene que mantenerlas “encendidas” y, al mismo tiempo, inhibirlas continuamente para no mezclarlas. Ese control constante genera una carga cognitiva extra que, sostenida a lo largo de los años, se asocia con un refuerzo de la reserva cognitiva.

El uso prolongado de varios idiomas fortalecería las redes cerebrales de control ejecutivo, atención y memoria a través de un desafío cognitivo continuo. Justamente estas mismas redes son las más vulnerables al declive de la edad, al deterioro cognitivo leve y a la demencia, lo que sitúa al multilingüismo como un posible factor de protección de la salud cerebral a lo largo de la vida.

[Anushka @Kulfei]

kasa tari hotey has no translations to any other language

[Blake Burge @blakeaburge]

A friend of mine used to say: “Show up on time, with a good attitude, and do what you said you’d do. That’s it. That’s 90% of winning in life." The older I get, the more I realize just how right he was.

[Mike Pesca @pescami]

Ordinary Abundance. This is the greatest thing I've read in weeks. No kidding. This should win a Pulitzer in a category not yet invented. https://t.co/iqnoACNt6z

[Physics In History @PhysInHistory]

The vacuum has structure. Empty space carries fields, fluctuations, and a nonzero energy we can measure the effects of and cannot compute.

[kian 🏴󠁧󠁢󠁥󠁮󠁧󠁿 @kianlufc]

That emergency alert is going to cause some serious office small talk around the country on Monday morning

[Nishi @nishimaru30]

You really didn’t need to buzz 5 times 😭

[Politics UK @PolitlcsUK]

🚨 BREAKING: The Government will shortly issue an emergency alert to all phones in England and Wales warning them not to light anything that can cause a wildfire

[˚꩜｡ @yappingwmogu143]

in another life i am a literature student in oxford.

[Jorge Bravo Abad @bravo_abad]

Can machine learning do more than interpolate within known chemistry and actually help recover reactions that previously did not work?

Paulo Neves and coauthors combine high-throughput experimentation with active learning to tackle Buchwald–Hartwig coupling, a key C–N bond-forming reaction in pharmaceutical synthesis where predictive models often struggle on genuinely new substrates and conditions.

The applied ML idea is strongly data-centric. The authors standardize multiple heterogeneous reaction datasets, generate 11,300 new high-quality reactions, and then use active learning to decide where new experiments should expand chemical space. Their selection strategy explores a virtual space of 64 million reactions using model uncertainty together with chemical diversity, rather than simply collecting more data.

The most interesting ML result is that diversity matters more than scale. Their Compound-Reaction Diversity Score, which captures coverage of reactant pairs and reagent combinations, correlates with out-of-distribution ROC AUC at Pearson r=0.79, compared with r=0.60 for dataset size alone. And with the curated dataset, even a relatively simple random forest exceeds 0.90 ROC AUC for high-confidence OOD predictions.

The chemistry result is stronger because it is experimentally validated. The model ranked 1,092 reagent combinations for 11 substrate pairs that had previously failed. Only the top four recommendations per pair were tested. 27 of 44 reactions succeeded, yielding viable conditions for 10 of the 11 previously unsuccessful substrate pairs.

The authors then pushed further into entirely unseen chemistry. From a virtual space of about 2.46x10^11 possible Buchwald–Hartwig reactions, high-confidence model recommendations achieved a 33% success rate versus 19% for the baseline set, corresponding to a 1.64× enrichment.

The broader AI-for-Science lesson is important: when extrapolation matters and experiments are expensive, the decisive question may not be “which model is more sophisticated?” but which experiments make the dataset more informative? Active learning makes data acquisition itself part of the learning algorithm.

Paper: Neves et al., Nature Computational Science (2026), CC BY 4.0 | https://t.co/OlzETpZXfv

[Philippe Schwaller (he/him) @pschwllr]

My team (@SchwallerGroup) has been cooking something big, and I'm really proud of it. Excited to share "Strategy-first synthesis planning for complex natural products"!

We are entering the era of agentic synthesis planning, and for the first time I see expert synthetic chemists excited about (some, not all) AI-generated routes, a stark contrast to anything we worked on before.

Solving routes (the field's default) does not mean solving them elegantly or strategically. That is what we've been working on. And natural products are a space typically out of reach for other ML-based planning tools.

Quick summary:

SynthEx is an agentic planner: a language model writes disconnections as atom-level graph edits rather than picking from a template library. A multi-agent system runs the loop: one agent proposes competing strategies per target, each anchored on a key disconnection, another expands them into full routes, and a critic-editor pair simulates each step forward and repairs what would fail.

→ On 1,098 complex natural products with no reported total synthesis, a near-exhaustive AiZynthFinder run solves 13.8%. SynthEx reaches 63.9%, and the margin widens as molecules get more complex.
→ In blinded review, 10 expert chemists gave 1,040 ratings over 148 key steps. SynthEx's steps were indistinguishable from published human ones on feasibility, elegance and overall quality, and raters could not tell machine from human (AUC 0.48).

Case studies highlight:

→ Okaramine M: it recovers the strategic logic of an expert route published after the model's training cut-off.

→ Melonine: it converges on exactly the key disconnection an expert group had committed lab effort to. That attempt failed for conformational reasons; SynthEx reaches it by a variant the same group judges more likely to succeed.

→ Chanoclavine to Lysergol: an open problem with no reported route. From two structures alone, it proposed a Hofmann-Löffler-Freytag sequence with a 1,6-HAT.

Huge collaborative project with the Zhu, Wipf and Njardarson groups, whose feedback shaped this throughout.

All the authors: Daniel Armstrong (@d_armstr )*, Xuan-Vu Nguyen* (@XuanVuNguyen18), Octavian Susanu, Gabriel Gibberd, Théo Neukomm, Taddäus Strunden, Dan Forster, Morgane Delattre, Shawn Teh, Clément Rols, John Federice, Hayden Leatherwood, Lavelle Barnes, Maarten Dobbelaere, Peter Wipf (@wipf_group), Jon Njardarson, Jieping Zhu, Philippe Schwaller (*equal)

Grateful for funding from the Swiss National Science Foundation SNSF (@snsf_ch) via @NCCR_Catalysis (225147) and grant 214915; the MSCA Doctoral Networks Explainable AI for Molecules - @AiChemist_DN MSCA DN Horizon Europe and LowDataML for Sustainable Chemical Sciences; and Intel and Merck Group via AWASES.

Made feasible by @Google Cloud Research Credits for the @GeminiApp Academic Program, with support from Digital Schooling. The USD 50k award I received in early 2026 let us build and run this without worrying about API costs. Backbone: Gemini 3.1 Pro.

We link related reactions via the @ElsevierConnect Reaxys R&D collaboration network.

-> SynthAtlas (explore all 3,243 routes, and leave comments): https://t.co/pW98YtmOTH

-> Preprint: https://t.co/95gbpOCFe8 (SynthEx method)

-> Code: https://t.co/MimVyj3tp2 - set Watch to be notified when it lands under Apache 2.0

[Dev Latimer @DevinRN88]

don't know the secret to happiness, but I've never been sad in a bookstore.

[Jay Alto @theJayAlto]

we need more people who are unapologetically in the pursuit of greatness

https://t.co/my8Erz5dZJ

[Prompter @PromptLLM]

The quality of your life is measured by your tolerance for uncertainty

[Rowan @RowanSci]

Rowan's NMR workflow now computes and visualizes J couplings:

[Mathieu @miniapeur]

More

[Nishi @nishimaru30]

@miniapeur they really cute!!!