Sign in

Viktor Zaverkin

@viktorzaverkin.bsky.social
70 followers 120 following 29 posts

Research Scientist @ NEC Labs Europe, Ph.D. in Theoretical Chemistry @ SimTech & @unistuttgart.bsky.social, ML/DL for Chemistry & Materials Science

PostsRepliesMedia
Viktor Zaverkin @viktorzaverkin.bsky.social · 24/08/2025
For more details about the work on learning uniformly accurate interatomic potentials from scratch I'll present in B1.36: 📄 Paper: www.nature.com/articles/s41... 💻 Code: github.com/nec-research...
nature.com
Uncertainty-biased molecular dynamics for learning uniformly accurate interatomic potentials - npj Computational Materials
npj Computational Materials - Uncertainty-biased molecular dynamics for learning uniformly accurate interatomic potentials
000
Viktor Zaverkin @viktorzaverkin.bsky.social · 24/08/2025
Or stop by poster B1.36 (Thu, Aug 28)! #PsiK2025 #AI4Science
100
Viktor Zaverkin @viktorzaverkin.bsky.social · 24/08/2025
I’ll be at the Psi-k conference next week! Let’s chat about ML potentials, uncertainty quantification (ensemble-free, gradient-based: Laplace approx., NTKs, batch selection, …), uncertainty-biased MD, message-passing architectures, particle-mesh long-range methods, etc.
100
Viktor Zaverkin @viktorzaverkin.bsky.social · 15/08/2025
🧵 TL;DR: ✅ Benchmark metrics improve with model size and electrostatics ❌ These gains don't always translate to improved simulation outcomes ⚠️ Training data & evaluation practices remain key bottlenecks 📄Preprint: arxiv.org/abs/2508.10841 💻Code: github.com/nec-research... -/
arxiv.org
Performance of universal machine-learned potentials with explicit long-range interactions in biomolecular simulations
Universal machine-learned potentials promise transferable accuracy across compositional and vibrational degrees of freedom, yet their application to biomolecular simulations remains underexplored. Thi...
000
Viktor Zaverkin @viktorzaverkin.bsky.social · 15/08/2025
With @matheusfferraz.bsky.social, @falesiani.bsky.social, @mniepert.bsky.social 15/
100
Viktor Zaverkin @viktorzaverkin.bsky.social · 15/08/2025
Moreover, simulation results are sensitive to training data composition. E.g., water density predictions depend on whether NaCl-water clusters were in the training set: compare ICTP-LR(M) vs. ICTP-LR(M)*. Legend: Solid green - ICTP-LR(M); dashed green - ICTP-LR(M)*. 14/
100
Viktor Zaverkin @viktorzaverkin.bsky.social · 15/08/2025
Without DFT-level simulations or other baselines, it is difficult to assess to what extent universal ML potentials improve on classical FFs in realistic biomolecular settings. While their qualitative advantages are often evident, quantitative validation remains challenging. 13/
100
Viktor Zaverkin @viktorzaverkin.bsky.social · 15/08/2025
These results highlight the limitations of current evaluation practices. 12/
100
Viktor Zaverkin @viktorzaverkin.bsky.social · 15/08/2025
In Trp-cage, simulations with explicit long-range electrostatics exhibit greater conformational variability. However, the origin of these effects remains unclear without DFT-level simulations. 11/
100
Viktor Zaverkin @viktorzaverkin.bsky.social · 15/08/2025
For Crambin, no significant differences are observed for the vibrational spectrum. 10/
100
Viktor Zaverkin @viktorzaverkin.bsky.social · 15/08/2025
For Ala3, larger models better reproduce experimental J-couplings. 9/
100
Viktor Zaverkin @viktorzaverkin.bsky.social · 15/08/2025
For water and NaCl-water mixtures: - Larger models don't consistently outperform smaller ones - Increasing model size doesn't yield systematic convergence - Explicit electrostatics shifts density predictions from overestimation to underestimation, without consistent gains. 8/
100
Viktor Zaverkin @viktorzaverkin.bsky.social · 15/08/2025
BUT: These improvements do not consistently translate into more accurate physical observables in simulations. Densities, radial distribution functions, and conformational ensembles show inconsistent trends with model size and long-range electrostatics. 7/
100
Viktor Zaverkin @viktorzaverkin.bsky.social · 15/08/2025
As expected, benchmark metrics (e.g., energy & force RMSEs) systematically improve with increasing model size and the inclusion of explicit long-range interactions. 6/
100
Viktor Zaverkin @viktorzaverkin.bsky.social · 15/08/2025
We use DIMOS for our simulations: 📄Preprint: arxiv.org/abs/2503.20541 💻Code: github.com/nec-research... 5/
100
Viktor Zaverkin @viktorzaverkin.bsky.social · 15/08/2025
We assess the impact of model size, dataset composition, and explicit long-range electrostatics across: 📊 Benchmark datasets 💧 Pure liquid water 🧂 NaCl-water mixtures 🧬 Small peptides (blocked and cationic Ala3) 🧪 Small proteins (Trp-cage, Crambin) 4/
100
Viktor Zaverkin @viktorzaverkin.bsky.social · 15/08/2025
DFT-level simulations and other high-quality baselines are unavailable or infeasible for biomolecular systems. A more reliable evaluation should consider how model expressivity (model size, explicit long-range interactions) affects prediction errors and simulation results. 3/
100
Viktor Zaverkin @viktorzaverkin.bsky.social · 15/08/2025
📄 Preprint: arxiv.org/abs/2508.10841 💻 Code: github.com/nec-research... 2/
arxiv.org
Performance of universal machine-learned potentials with explicit long-range interactions in biomolecular simulations
Universal machine-learned potentials promise transferable accuracy across compositional and vibrational degrees of freedom, yet their application to biomolecular simulations remains underexplored. Thi...
100
Viktor Zaverkin @viktorzaverkin.bsky.social · 15/08/2025
🚨 New preprint: How well do universal ML potentials perform in biomolecular simulations under realistic conditions? There's growing excitement around ML potentials trained on large datasets. But do they deliver in simulations of biomolecular systems? It’s not so clear. 🧵 1/
151
Viktor Zaverkin @viktorzaverkin.bsky.social · 25/06/2025
📎Paper: pubs.acs.org/doi/abs/10.1... (but check also pubs.acs.org/doi/abs/10.1...) 💻Code(s): gitlab.com/zaverkin_v/g... (TensorFlow), github.com/nec-research... (PyTorch), and github.com/apax-hub/apax (JAX)
pubs.acs.org
Gaussian Moments as Physically Inspired Molecular Descriptors for Accurate and Scalable Machine Learning Potentials
Machine learning techniques allow a direct mapping of atomic positions and nuclear charges to the potential energy surface with almost ab initio accuracy and the computational efficiency of empirical ...
000
Viktor Zaverkin @viktorzaverkin.bsky.social · 25/06/2025
Many thanks to everyone who has read, cited, or built on it. I hope it continues to be helpful!
100
Viktor Zaverkin @viktorzaverkin.bsky.social · 25/06/2025
We proposed using full tensor contractions to construct many-body features, thereby avoiding expensive sums over triplets, quadruplets, and so on. I am thrilled to see that similar ideas are now an integral part of state-of-the-art architectures, such as MACE, CACE, and so on. 💪
100
Viktor Zaverkin @viktorzaverkin.bsky.social · 25/06/2025
📈My first PhD paper just reached 100 citations, which is a small but very special milestone for me! Our paper introduces Gaussian moments as molecular descriptors and uses them to build ML potentials with an impressive balance between accuracy and computational efficiency.
130
Reposted by Viktor Zaverkin
Chemical Sciences (M.Sc.) | University of Stuttgart @chemicalsciences.bsky.social · 03/02/2025
🚀 Apply Now: International Master's Chemical Sciences! 🌍 The application portal for the English-conducted M.Sc. Chemical Sciences at @unistuttgart.bsky.social are officially open! 🔬✨ 👉 Visit our program website for further details: www.uni-stuttgart.de/en/study/stu...
uni-stuttgart.de
Chemical Sciences M.Sc. for prospective students | Study program | University of Stuttgart
Information for prospective students: Chemical Sciences at the University of Stuttgart: application, admission, requirements.
133
Reposted by Viktor Zaverkin
David Holzmüller @dholzmueller.bsky.social · 12/12/2024
I'll present our paper in the afternoon poster session at 4:30pm - 7:30 pm in East Exhibit Hall A-C, poster 3304!
032
Reposted by Viktor Zaverkin
Daniel Musekamp @danielmusekamp.bsky.social · 11/12/2024
Neural surrogates can accelerate PDE solving but need expensive ground-truth training data. Can we reduce the training data size with active learning (AL)? In our NeurIPS D3S3 poster, we introduce AL4PDE, an extensible AL benchmark for autoregressive neural PDE solvers. 🧵
1123
Viktor Zaverkin @viktorzaverkin.bsky.social · 11/12/2024
Join us today at #NeurIPS2024 for our poster presentation: Higher-Rank Irreducible Cartesian Tensors for Equivariant Message Passing 🗓️ When: Wed, Dec 11, 11 a.m. – 2 p.m. PST 📍 Where: East Exhibit Hall A-C, Poster #4107 #MachineLearning #InteratomicPotentials #Equivariance #GraphNeuralNetworks
051
Viktor Zaverkin @viktorzaverkin.bsky.social · 11/12/2024
@falesiani.bsky.social
010
Viktor Zaverkin @viktorzaverkin.bsky.social · 08/12/2024
@takashimal.bsky.social
100
Viktor Zaverkin @viktorzaverkin.bsky.social · 08/12/2024
Good question! There is a certain connection between Cartesian and geometric products through those operations: geometric product for an n-dimensional vector could be seen as an "outer" product (up to n-rank tensors) + a "contraction". However, I don't know of any systematic study of the relation.
000
Reposted by Viktor Zaverkin
Chaitanya K. Joshi @chaitjo.bsky.social · 07/12/2024
My take: Folks from deep learning (I am guilty too): super excited about methods, modelling, make it look mathy, our number is bold in the table that everyone re-uses for 5 years... Downstream users of tools in drug discovery: Oh god - is that how you evaluated? *Surprised Pikachu meme*
2173
Viktor Zaverkin @viktorzaverkin.bsky.social · 06/12/2024
A big shoutout to the incredible team behind this work: Francesco Alesiani, Takashi Maruyama, @federico-errica.bsky.social, Henrik Christiansen, Makoto Takamoto, Nicolas Weber, and @mniepert.bsky.social. Huge thanks to NEC Labs Europe, @unistuttgart.bsky.social , and SimTech for their support!
140
Viktor Zaverkin @viktorzaverkin.bsky.social · 06/12/2024
📣 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...
2132