The Open Molecules 2025 dataset is out! With >100M gold-standard ωB97M-V/def2-TZVPD calcs of biomolecules, electrolytes, metal complexes, and small molecules, OMol is by far the largest, most diverse, and highest quality molecular DFT dataset for training MLIPs ever made 1/N
May 14, 2025 at 8:52 PM
The Open Molecules 2025 dataset is out! With >100M gold-standard ωB97M-V/def2-TZVPD calcs of biomolecules, electrolytes, metal complexes, and small molecules, OMol is by far the largest, most diverse, and highest quality molecular DFT dataset for training MLIPs ever made 1/N
You should take a look at this if you want to know how to use Cartesian (instead of spherical) tensors for building equivariant MLIPs.
📣 Can we go beyond state-of-the-art message-passing models based on spherical tensors such as #MACE and #NequIP?
Our #NeurIPS2024 paper explores higher-rank irreducible Cartesian tensors to design equivariant #MLIPs.
Paper: arxiv.org/abs/2405.14253
Code: github.com/nec-research...
Our #NeurIPS2024 paper explores higher-rank irreducible Cartesian tensors to design equivariant #MLIPs.
Paper: arxiv.org/abs/2405.14253
Code: github.com/nec-research...
December 6, 2024 at 2:57 PM
You should take a look at this if you want to know how to use Cartesian (instead of spherical) tensors for building equivariant MLIPs.
Most of MLIPs dont distinguish between spin states, making them unsuitable for open-shell chemistry. We present AIMNet2-NSE (Neural Spin-charge Equilibration), MLIP that incorporates spin-charge equilibration for systems with arbitrary charge and spin. #compchem #skychem
chemrxiv.org/engage/chemr...
chemrxiv.org/engage/chemr...
August 4, 2025 at 4:47 PM
Most of MLIPs dont distinguish between spin states, making them unsuitable for open-shell chemistry. We present AIMNet2-NSE (Neural Spin-charge Equilibration), MLIP that incorporates spin-charge equilibration for systems with arbitrary charge and spin. #compchem #skychem
chemrxiv.org/engage/chemr...
chemrxiv.org/engage/chemr...
New Preprint dropped. LAMMPS + ANI, a super fast implementation of MLIPs. Highly parallel (> 1000 GPUS) and much faster than anything else out there ! Work done by the amazing @ignaciopickering.bsky.social, @nickterrel.bsky.social , and Jinze (Richard) Xue. chemrxiv.org/engage/chemr...
LAMMPS-ANI: Large Scale Molecular Dynamics Simulations
with ANI Neural Network Potential
Machine Learning Interatomic Potentials (MLIPs), trained with Quantum Mechanics data, can model
potential energy surfaces for molecular systems with very high accuracy and extreme speedups compared
to...
chemrxiv.org
January 5, 2026 at 9:11 PM
New Preprint dropped. LAMMPS + ANI, a super fast implementation of MLIPs. Highly parallel (> 1000 GPUS) and much faster than anything else out there ! Work done by the amazing @ignaciopickering.bsky.social, @nickterrel.bsky.social , and Jinze (Richard) Xue. chemrxiv.org/engage/chemr...
Exciting @cmuchemistry.bsky.social event: research updates and poster presentations by the 2nd year graduate students. Two are from our lab: 1)extension of AIMNet2 MLIP towards macrocyles 2) protein-ligand interactions from physics-informed MLIPs. Good luck, everyone 🍀 #chemsky #compchem
February 10, 2025 at 9:37 PM
Exciting @cmuchemistry.bsky.social event: research updates and poster presentations by the 2nd year graduate students. Two are from our lab: 1)extension of AIMNet2 MLIP towards macrocyles 2) protein-ligand interactions from physics-informed MLIPs. Good luck, everyone 🍀 #chemsky #compchem
I'll be at NeurIPS. I'm still looking for an intern: Generative modeling for discrete & continuous variables that optimize reward functions, or doing model distillation on big MLIPs, or your cool idea! Contact me on whova or here!
December 8, 2024 at 3:28 PM
I'll be at NeurIPS. I'm still looking for an intern: Generative modeling for discrete & continuous variables that optimize reward functions, or doing model distillation on big MLIPs, or your cool idea! Contact me on whova or here!
New @chemrxiv.org preprint: AIMNet2(Score), a label-free way to score protein-ligand complexes. No experimental affinity data. Just QM-level physics from AIMNet2 MLIPs: interaction energy + ligand desolvation + conformational strain, on a single bound structure. #compchem #chemsky🧵
July 24, 2026 at 8:05 AM
New @chemrxiv.org preprint: AIMNet2(Score), a label-free way to score protein-ligand complexes. No experimental affinity data. Just QM-level physics from AIMNet2 MLIPs: interaction energy + ligand desolvation + conformational strain, on a single bound structure. #compchem #chemsky🧵
Had a fantastic time in Kyoto last week! It was a pleasure to present @Tomasz's work on supramolecular catalysis (in collaboration with the Lusby group) at #ISMSC2025 and @juraskova-ver.bsky.social and Hanwen's work on reactive MLIPs at #MQM2025
June 3, 2025 at 5:46 AM
Had a fantastic time in Kyoto last week! It was a pleasure to present @Tomasz's work on supramolecular catalysis (in collaboration with the Lusby group) at #ISMSC2025 and @juraskova-ver.bsky.social and Hanwen's work on reactive MLIPs at #MQM2025
You can check out mlip-playground.bragitoff.com to play around with over 30 universal MLIPs from FairChem, MACE, MatterSim, SevenNet and Orb.
MLIP Playground - Run, Test and Benchmark MLIPs
Run, Test and Benchmark over 20 foundation models
mlip-playground.bragitoff.com
November 26, 2025 at 7:09 PM
You can check out mlip-playground.bragitoff.com to play around with over 30 universal MLIPs from FairChem, MACE, MatterSim, SevenNet and Orb.
📣 Can we go beyond state-of-the-art message-passing models based on spherical tensors such as #MACE and #NequIP?
Our #NeurIPS2024 paper explores higher-rank irreducible Cartesian tensors to design equivariant #MLIPs.
Paper: arxiv.org/abs/2405.14253
Code: github.com/nec-research...
Our #NeurIPS2024 paper explores higher-rank irreducible Cartesian tensors to design equivariant #MLIPs.
Paper: arxiv.org/abs/2405.14253
Code: github.com/nec-research...
December 6, 2024 at 2:45 PM
📣 Can we go beyond state-of-the-art message-passing models based on spherical tensors such as #MACE and #NequIP?
Our #NeurIPS2024 paper explores higher-rank irreducible Cartesian tensors to design equivariant #MLIPs.
Paper: arxiv.org/abs/2405.14253
Code: github.com/nec-research...
Our #NeurIPS2024 paper explores higher-rank irreducible Cartesian tensors to design equivariant #MLIPs.
Paper: arxiv.org/abs/2405.14253
Code: github.com/nec-research...
What do transition state search, geometry relaxation, zero point energy corrections, and extrema classification have in common? They all require Hessians!
The problem is, accurate Hessians are really expensive, even with MLIPs. We say, just shoot 'em from the HIP! 🤠
[1/4]
The problem is, accurate Hessians are really expensive, even with MLIPs. We say, just shoot 'em from the HIP! 🤠
[1/4]
October 8, 2025 at 5:47 PM
What do transition state search, geometry relaxation, zero point energy corrections, and extrema classification have in common? They all require Hessians!
The problem is, accurate Hessians are really expensive, even with MLIPs. We say, just shoot 'em from the HIP! 🤠
[1/4]
The problem is, accurate Hessians are really expensive, even with MLIPs. We say, just shoot 'em from the HIP! 🤠
[1/4]
🎓 Applications are open for the IMAT Centre for Doctoral Training at @oxfordchemistry.bsky.social & @oxfordmaterials.bsky.social!
Among many exciting topics, have a look at our projects P10 and P13, both combining MLIPs and #compchem with close experimental collaborations: imatcdt.chem.ox.ac.uk
Among many exciting topics, have a look at our projects P10 and P13, both combining MLIPs and #compchem with close experimental collaborations: imatcdt.chem.ox.ac.uk
Home | IMAT CDT
imatcdt.chem.ox.ac.uk
November 11, 2025 at 11:29 AM
🎓 Applications are open for the IMAT Centre for Doctoral Training at @oxfordchemistry.bsky.social & @oxfordmaterials.bsky.social!
Among many exciting topics, have a look at our projects P10 and P13, both combining MLIPs and #compchem with close experimental collaborations: imatcdt.chem.ox.ac.uk
Among many exciting topics, have a look at our projects P10 and P13, both combining MLIPs and #compchem with close experimental collaborations: imatcdt.chem.ox.ac.uk
New paper on the structure of amorphous arsenic together with @vlderinger.bsky.social 's group! Automated training of autoplex was used to train the MLIPs and my group contributed to the bonding analysis! 😃
www.arxiv.org/abs/2509.02484
#compchemsky
www.arxiv.org/abs/2509.02484
#compchemsky
www.arxiv.org
September 3, 2025 at 8:33 PM
New paper on the structure of amorphous arsenic together with @vlderinger.bsky.social 's group! Automated training of autoplex was used to train the MLIPs and my group contributed to the bonding analysis! 😃
www.arxiv.org/abs/2509.02484
#compchemsky
www.arxiv.org/abs/2509.02484
#compchemsky
A primer for non conservative (& rotationally unconstrained) MLIPs, and how to use them safely. Thanks @aihub.org for the space! aihub.org/2025/10/10/m...
Machine learning for atomic-scale simulations: balancing speed and physical laws - ΑΙhub
aihub.org
October 10, 2025 at 10:31 AM
A primer for non conservative (& rotationally unconstrained) MLIPs, and how to use them safely. Thanks @aihub.org for the space! aihub.org/2025/10/10/m...
🚀 We asked: What if the time to set up and run the best Machine-Learned Interatomic Potentials (MLIPs) took seconds, not days?
Today, we release the MLIP Garden v0.1.
What you can do now:
- Experiment ~instantly
- Scale deployments on experimental NSF and Dept of Energy systems
Today, we release the MLIP Garden v0.1.
What you can do now:
- Experiment ~instantly
- Scale deployments on experimental NSF and Dept of Energy systems
August 20, 2025 at 7:27 PM
🚀 We asked: What if the time to set up and run the best Machine-Learned Interatomic Potentials (MLIPs) took seconds, not days?
Today, we release the MLIP Garden v0.1.
What you can do now:
- Experiment ~instantly
- Scale deployments on experimental NSF and Dept of Energy systems
Today, we release the MLIP Garden v0.1.
What you can do now:
- Experiment ~instantly
- Scale deployments on experimental NSF and Dept of Energy systems
I don't think it will replace anything. Just how LDA, GGAs, and meta-GGAs can coexist, FFs, MLIPs, and DFT will coexist. MLIPs will just take over some of the tasks where FFs aren't accurate enough and DFT isn't fast enough.
September 10, 2025 at 6:22 PM
I don't think it will replace anything. Just how LDA, GGAs, and meta-GGAs can coexist, FFs, MLIPs, and DFT will coexist. MLIPs will just take over some of the tasks where FFs aren't accurate enough and DFT isn't fast enough.
I'm hiring postdocs @berkeleylab.lbl.gov to drive cutting-edge research involving MLIPs, high-throughput workflows, chemical reaction networks, generative models, and open-source software dev. Full position description + application here: forms.gle/zePBZDmciXez... #Chempostdoc #AI4Science
forms.gle
November 4, 2025 at 9:16 PM
I'm hiring postdocs @berkeleylab.lbl.gov to drive cutting-edge research involving MLIPs, high-throughput workflows, chemical reaction networks, generative models, and open-source software dev. Full position description + application here: forms.gle/zePBZDmciXez... #Chempostdoc #AI4Science
In fairness, I suspect few/non of these MLIPs were trained on dimers and that hence the dimers here are only used as a test for the universality of the MLIPs, a task they only do so so at.
November 17, 2024 at 10:22 PM
In fairness, I suspect few/non of these MLIPs were trained on dimers and that hence the dimers here are only used as a test for the universality of the MLIPs, a task they only do so so at.
Do you use machine-learned interatomic potentials (MLIPs) on HPC clusters? Do you have a graveyard of condo environments to manage their incompatible dependencies?
If so, we should talk.
We built Rootstock to make it trivial to swap MLIPs when running atomistic simulation jobs with ASE/LAMMPS.
If so, we should talk.
We built Rootstock to make it trivial to swap MLIPs when running atomistic simulation jobs with ASE/LAMMPS.
March 18, 2026 at 3:23 PM
Do you use machine-learned interatomic potentials (MLIPs) on HPC clusters? Do you have a graveyard of condo environments to manage their incompatible dependencies?
If so, we should talk.
We built Rootstock to make it trivial to swap MLIPs when running atomistic simulation jobs with ASE/LAMMPS.
If so, we should talk.
We built Rootstock to make it trivial to swap MLIPs when running atomistic simulation jobs with ASE/LAMMPS.
It goes deeper - some people prefer NNIP, though technically there is a difference (all NNIPs are MLIPs, but not all MLIPs are NNIPs)
November 14, 2024 at 7:48 AM
It goes deeper - some people prefer NNIP, though technically there is a difference (all NNIPs are MLIPs, but not all MLIPs are NNIPs)
No day goes by without a new universal #ML potential. But how different they really are? Sanggyu and Sofiia tried to give a quantitative answer by comparing the reconstruction errors between their latent-space features. If you are curious, check out the #preprint arxiv.org/html/2512.05...
December 9, 2025 at 7:16 AM
No day goes by without a new universal #ML potential. But how different they really are? Sanggyu and Sofiia tried to give a quantitative answer by comparing the reconstruction errors between their latent-space features. If you are curious, check out the #preprint arxiv.org/html/2512.05...
Machine learning potentials enable near-quantum accuracy at large scales. They bridge gaps between costly ab initio and simpler force fields. With improving data, methods, & GPU-driven speed, MLIPs promise a new era of faster, more accurate molecular simulations www.sciencedirect.com/science/arti...
March 4, 2025 at 12:52 PM
Machine learning potentials enable near-quantum accuracy at large scales. They bridge gaps between costly ab initio and simpler force fields. With improving data, methods, & GPU-driven speed, MLIPs promise a new era of faster, more accurate molecular simulations www.sciencedirect.com/science/arti...
Yup, and with MLIPs being very accessible, I'm seeing more in that space too. The number of adsorption studies one can dream up is literally infinite... 😭
Before, I was okay with accepting "technically sound but provides little value." Now there is going to be far more of it, for little gain.
Before, I was okay with accepting "technically sound but provides little value." Now there is going to be far more of it, for little gain.
September 22, 2026 at 5:15 PM
Yup, and with MLIPs being very accessible, I'm seeing more in that space too. The number of adsorption studies one can dream up is literally infinite... 😭
Before, I was okay with accepting "technically sound but provides little value." Now there is going to be far more of it, for little gain.
Before, I was okay with accepting "technically sound but provides little value." Now there is going to be far more of it, for little gain.
If you used our ANI MLIPs, you probably used our TorchANI library. Now, new and improved version 2.0. Use it, enjoy it, break it, let us know what you did or tried to do with it. doi.org/10.1021/acs.jcim.5c01853
@ignaciopickering.bsky.social @nickterrel.bsky.social @khuddleston.bsky.social
@ignaciopickering.bsky.social @nickterrel.bsky.social @khuddleston.bsky.social
TorchANI 2.0: An Extensible, High-Performance Library for the Design, Training, and Use of NN-IPs
In this work, we introduce TorchANI 2.0, a significantly improved version of the free and open source TorchANI software package for training and evaluation of ANI (ANAKIN-ME) deep learning models. Tor...
pubs.acs.org
October 17, 2025 at 9:32 PM
If you used our ANI MLIPs, you probably used our TorchANI library. Now, new and improved version 2.0. Use it, enjoy it, break it, let us know what you did or tried to do with it. doi.org/10.1021/acs.jcim.5c01853
@ignaciopickering.bsky.social @nickterrel.bsky.social @khuddleston.bsky.social
@ignaciopickering.bsky.social @nickterrel.bsky.social @khuddleston.bsky.social