Olexandr Isayev 🇺🇦 🇺🇸

Olexandr Isayev 🇺🇦 🇺🇸

@olexandr.bsky.social

Chemistry professor at CMU. Connecting chemical sciences with AI #MachineLearning and automated experimentation. #tarheels fan. Care: #design, #photography #Ukraine #cats🐈 Rants are mine

4.5K followers1.1K following889 posts

Olexandr Isayev 🇺🇦 🇺🇸 reposted

CMU Mellon College of Science
@cmuscience.bsky.social

Join MCS and @cmuchemistry.bsky.social for the Pople Lecture in Theoretical & Computational Chemistry on Oct. 7. Martin Head-Gordon, professor at U.C. Berkeley, will present "Density Functional Theory in Chemistry: Past, Present and Future." Register: www.givecampus.com/schools/Carn... #ChemSky

cmu.eduPople Lecture Connects Nobel Past With Chemistry’s FutureThe Royal Swedish Academy of Sciences will announce the winner or winners of the 2026 Nobel Prize in Chemistry on Oct. 7. Later that day, Carnegie Mellon University will gather to celebrate the legacy...
Olexandr Isayev 🇺🇦 🇺🇸
@olexandr.bsky.social

About to board my plane to Riga! Honored to speak at the Enamine 🇺🇦 Drug Discovery Conference. Looking forward to seeing old friends and making new ones. In my talk I will feature CMU AI Founry - our autonomous science lab and advances with AI and robotic experimentation 🤖

Olexandr Isayev 🇺🇦 🇺🇸
@olexandr.bsky.social

Faculty position (open rank, open area of chemistry) at CMU! Come to be my colleague! Let me know if you have any questions about this position. #compchem #chemsky #chemfolks #chemjobs #academicjobs

Chemistry @ Carnegie Mellon University@cmuchemistry.bsky.social

Come work with us at CMU! Applications are being accepted for multiple tenure-track faculty positions with an expected Fall 2027 start. Applications should be submitted online through Interfolio apply.interfolio.com/191917 . Review of applications will begin September 30, 2026.

apply.interfolio.com Apply - Interfolio {{$ctrl.$state.data.pageTitle}} - Apply - Interfolio

Olexandr Isayev 🇺🇦 🇺🇸 reposted

Chemical Science
@chemicalscience.rsc.org

Professor Olexandr Isayev from Carnegie Mellon University, USA will be speaking at the Chemical Science Symposium on the 29-30th October! They will be speaking on accelerating molecular design with autonomous AI agents. Register for the event here: rsc.li/chemscisymp2026 #ChemSky #CompChemSky

Olexandr Isayev 🇺🇦 🇺🇸
@olexandr.bsky.social

We built KIN66, a kinase benchmark of ~1,000-atom active sites at three DFT levels. AIMNet2(2025) reproduces its B97-3c reference to 3 kcal/mol RMSE and transfers to systems larger than anything in training. One MLIP, DFT-level interaction energies at protein-ligand scale.

Olexandr Isayev 🇺🇦 🇺🇸
@olexandr.bsky.social

On FEP+ series and HiQBind co-crystals, it ranks affinities on par with or ahead of docking, SQM, and deep-learning baselines, with no affinity labels. On raw docked poses, everything ties, which points to the ceiling on pose fidelity, not scoring physics. chemrxiv.org/doi/abs/10.2...

chemrxiv.orgLabel-Free, Quantum-Mechanically Informed Scoring of Protein–Ligand Complexes with Machine-Learned Interatomic Potentials | ChemRxivVirtual chemical libraries now exceed billions of compounds, placing joint demands on the scoring tools used to prioritize candidates for accuracy and scalability. Supervised affinity models meet scal...
Olexandr Isayev 🇺🇦 🇺🇸
@olexandr.bsky.social

In structure-based drug design, ligand conformational strain helps answer a key question: is this bound pose chemically plausible? But the usual gas-phase shortcut can misestimate strain, especially for charged and highly polar ligands. #compchem

Olexandr Isayev 🇺🇦 🇺🇸
@olexandr.bsky.social

Key takeaway: ligand strain is not just a fixed molecular descriptor. It depends on the bound pose, binding pocket, charge, flexibility. That matters for pose assessment, lead optimization, selectivity design, and docking/free-energy workflows. #SBDD #CompChem #chemsky chemrxiv.org/doi/abs/10.2...

chemrxiv.orgEfficient and Accurate Ligand Strain Calculations in Solution with the AIMNet2 Neural Network Potential | ChemRxivConformational strain, the energy a ligand pays to adopt its bound conformation, contributes to binding affinity but is routinely misestimated. The standard estimate references a single gasphase energ...

Olexandr Isayev 🇺🇦 🇺🇸 reposted

Stand Up for Science!
@standupforscience.net

The NSF is under attack from all angles, and Trump’s firing of the entire NSB is another huge blow to American science. Sign your support for the NSB and email your Members of Congress at the link in our bio, or go here: Sign 👉 zurl.co/tjU4b Email 👉 zurl.co/CDsik #NSB #Science #NSF #Trump

Olexandr Isayev 🇺🇦 🇺🇸
@olexandr.bsky.social

Preprint alert!🚨We introduce PROBE: a framework for uncertainty quantification in #machinelearning interatomic potentials (#MLIP). Instead of the typical regression, PROBE frames #uncertainty as a simple binary question: is this prediction reliable or not? #comchem #skychem arxiv.org/abs/2605.00640

arxiv.orgKnowing when to trust machine-learned interatomic potentialsPrevailing machine-learned interatomic potential (MLIP) uncertainty-quantification methods rely on ensembles of independently trained backbones. These methods scale unfavorably with foundation-scale M...
View full thread
Olexandr Isayev 🇺🇦 🇺🇸
@olexandr.bsky.social

Across large held-out sets and two distinct MLIP architectures, PROBE outperforms ensemble disagreement as a binary reliability signal, with this advantage increasing with the expressiveness of the backbone representation, implying a favorable scaling trajectory toward foundation-scale MLIPs.

Olexandr Isayev 🇺🇦 🇺🇸
@olexandr.bsky.social

Preprint alert!🚨We introduce PROBE: a framework for uncertainty quantification in #machinelearning interatomic potentials (#MLIP). Instead of the typical regression, PROBE frames #uncertainty as a simple binary question: is this prediction reliable or not? #comchem #skychem arxiv.org/abs/2605.00640

arxiv.orgKnowing when to trust machine-learned interatomic potentialsPrevailing machine-learned interatomic potential (MLIP) uncertainty-quantification methods rely on ensembles of independently trained backbones. These methods scale unfavorably with foundation-scale M...
Olexandr Isayev 🇺🇦 🇺🇸
@olexandr.bsky.social

This simplification makes the problem tractable, and we show that a lightweight classifier trained on top of a frozen MLIP's internal representations is all you need. PROBE attention mechanism produces per-atom importance maps highlighting which atoms drove an unreliable prediction, at no cost.

Olexandr Isayev 🇺🇦 🇺🇸 reposted

Brendan
@chemscrapes.bsky.social

I don't know who needs to hear this today...but you are doing great! Just drink heaps of water and take a stroll. This has been a ChemScrapes community announcement! #chemsky #chemchat

Olexandr Isayev 🇺🇦 🇺🇸
@olexandr.bsky.social

• Works on realistic chemistry: up to ~50 atoms, including charged systems • Trained on ~1.6M xTB TS • Much larger than standard TS datasets. Good TS guesses are the bottleneck for mechanistic modeling, they determine if TS and IRC succeed. Toward scalable, automated reaction mechanism exploration 🚀

Olexandr Isayev 🇺🇦 🇺🇸 reposted

Shipwreck
@shipwreck75.bsky.social

Footage captured from passengers on an airliner above Florida, showing the exact moment the Artemis II mission lifts off.