arXiv subjects

Emine Kucukbenli

Publications and source records attributed to Emine Kucukbenli.

18 recordsLinked to original sources

arXiv2026

A fast summation method for the DFT-D3 dispersion correction

The DFT-D3 dispersion correction is routinely added to machine learning force fields (MLFFs) trained on dispersion-deficient functionals such as PBE. Its environment-dependent pair coefficients, however, break the atom-centered separability that fast summation methods require, forcing practitioners either to truncate D3 or to accept a substantial slowdown. We introduce FourierD3, a method that uses a functional low-rank decomposition to restore this separability and enable particle-mesh evaluation in $O(N\log N)$ time without a real-space cutoff on the dispersion sum.

physics.comp-ph↗

arXiv2026

La-Proteina: Atomistic Protein Generation via Partially Latent Flow Matching

Recently, many generative models for de novo protein structure design have emerged. Yet, only few tackle the difficult task of directly generating fully atomistic structures jointly with the underlying amino acid sequence. This is challenging, for instance, because the model must reason over side chains that change in length during generation. We introduce La-Proteina for atomistic protein design based on a novel partially latent protein representation: coarse backbone structure is modeled explicitly, while sequence and atomistic details are captured via per-residue latent variables of fixed dimensionality, thereby effectively side-stepping challenges of explicit side-chain representations. Flow matching in this partially latent space then models the joint distribution over sequences and full-atom structures. La-Proteina achieves state-of-the-art performance on multiple generation benchmarks, including all-atom co-designability, diversity, and structural validity, as confirmed through detailed structural analyses and evaluations. Notably, La-Proteina also surpasses previous models in atomistic motif scaffolding performance, unlocking critical atomistic structure-conditioned protein design tasks. Moreover, La-Proteina is able to generate co-designable proteins of up to 800 residues, a regime where most baselines collapse and fail to produce valid samples, demonstrating La-Proteina's scalability and robustness.

cs.LG↗

arXiv2026

Scaling Atomistic Protein Binder Design with Generative Pretraining and Test-Time Compute

Protein interaction modeling is central to protein design, which has been transformed by machine learning with applications in drug discovery and beyond. In this landscape, structure-based de novo binder design is cast as either conditional generative modeling or sequence optimization via structure predictors ("hallucination"). We argue that this is a false dichotomy and propose Proteina-Complexa, a novel fully atomistic binder generation method unifying both paradigms. We extend recent flow-based latent protein generation architectures and leverage the domain-domain interactions of monomeric computationally predicted protein structures to construct Teddymer, a new large-scale dataset of synthetic binder-target pairs for pretraining. Combined with high-quality experimental multimers, this enables training a strong base model. We then perform inference-time optimization with this generative prior, unifying the strengths of previously distinct generative and hallucination methods. Proteina-Complexa sets a new state of the art in computational binder design benchmarks: it delivers markedly higher in-silico success rates than existing generative approaches, and our novel test-time optimization strategies greatly outperform previous hallucination methods under normalized compute budgets. We also demonstrate interface hydrogen bond optimization, fold class-guided binder generation, and extensions to small molecule targets and enzyme design tasks, again surpassing prior methods. Code, models and new data will be publicly released.

cs.LG↗

arXiv2026

Fold-CP: A Context Parallelism Framework for Biomolecular Modeling

Understanding cellular machinery requires atomic-scale reconstruction of large biomolecular assemblies. However, predicting the structures of these systems has been constrained by hardware memory requirements of models like AlphaFold 3, imposing a practical ceiling of a few thousand residues that can be processed on a single GPU. Here we present NVIDIA BioNeMo Fold-CP, a context parallelism framework that overcomes this barrier by distributing the inference and training pipelines of co-folding models across multiple GPUs. We use the Boltz models as open source reference architectures and implement custom multidimensional primitives that efficiently parallelize both the dense triangular updates and the irregular, data-dependent pattern of window-batched local attention. Our approach achieves efficient memory scaling; for an N-token input distributed across P GPUs, per-device memory scales as $O(N^2/P)$, enabling the structure prediction of assemblies exceeding 30,000 residues on 64 NVIDIA B300 GPUs. We demonstrate the scientific utility of this approach through successful developer use cases: Fold-CP enabled the scoring of over 90% of Comprehensive Resource of Mammalian protein complexes (CORUM) database, as well as folding of disease-relevant PI4KA lipid kinase complex bound to an intrinsically disordered region without cropping. By providing a scalable pathway for modeling massive systems with full global context, Fold-CP represents a significant step toward the realization of a virtual cell.

q-bio.QM↗

arXiv2026

Augmenting Molecular Graphs with Geometries via Machine Learning Interatomic Potentials

Accurate molecular property predictions require 3D geometries, which are typically obtained using expensive methods such as density functional theory (DFT). Here, we attempt to obtain molecular geometries by relying solely on machine learning interatomic potential (MLIP) models. To this end, we first curate a large-scale molecular relaxation dataset comprising 3.5 million molecules and 300 million snapshots. Then MLIP pre-trained models are trained with supervised learning to predict energy and forces given 3D molecular structures. Once trained, we show that the pre-trained models can be used in different ways to obtain geometries either explicitly or implicitly. First, it can be used to obtain approximate low-energy 3D geometries via geometry optimization. While these geometries do not consistently reach DFT-level chemical accuracy or convergence, they can still improve downstream performance compared to non-relaxed structures. To mitigate potential biases and enhance downstream predictions, we introduce geometry fine-tuning based on the relaxed 3D geometries. Second, the pre-trained models can be directly fine-tuned for property prediction when ground truth 3D geometries are available. Our results demonstrate that MLIP pre-trained models trained on relaxation data can learn transferable molecular representations to improve downstream molecular property prediction and can provide practically valuable but approximate molecular geometries that benefit property predictions. Our code is publicly available at: https://github.com/divelab/AIRS/

physics.chem-ph↗

arXiv2026

Tensor Decomposition Networks for Fast Machine Learning Interatomic Potential Computations

$\rm{SO}(3)$-equivariant networks are the dominant models for machine learning interatomic potentials (MLIPs). The key operation of such networks is the Clebsch-Gordan (CG) tensor product, which is computationally expensive. To accelerate the computation, we develop tensor decomposition networks (TDNs) as a class of approximately equivariant networks in which CG tensor products are replaced by low-rank tensor decompositions, such as the CANDECOMP/PARAFAC (CP) decomposition. With the CP decomposition, we prove (i) a uniform bound on the induced error of $\rm{SO}(3)$-equivariance, and (ii) the universality of approximating any equivariant bilinear map. To further reduce the number of parameters, we propose path-weight sharing that ties all multiplicity-space weights across the $\mathcal{O}(L^3)$ CG paths into a single shared parameter set without compromising equivariance, where $L$ is the maximum angular degree. The resulting layer acts as a plug-and-play replacement for tensor products in existing networks, and the computational complexity of tensor products is reduced from $\mathcal{O}(L^6)$ to $\mathcal{O}(L^4)$. We evaluate TDNs on PubChemQCR, a newly curated molecular relaxation dataset containing 105 million DFT-calculated snapshots. We also use existing datasets, including OC20, and OC22. Results show that TDNs achieve competitive performance with dramatic speedup in computations. Our code is publicly available as part of the AIRS library (\href{https://github.com/divelab/AIRS/tree/main/OpenMol/TDN}{https://github.com/divelab/AIRS/}).

cs.LG↗

arXiv2025

Consistent Synthetic Sequences Unlock Structural Diversity in Fully Atomistic De Novo Protein Design

High-quality training datasets are crucial for the development of effective protein design models, but existing synthetic datasets often include unfavorable sequence-structure pairs, impairing generative model performance. We leverage ProteinMPNN, whose sequences are experimentally favorable as well as amenable to folding, together with structure prediction models to align high-quality synthetic structures with recoverable synthetic sequences. In that way, we create a new dataset designed specifically for training expressive, fully atomistic protein generators. By retraining La-Proteina, which models discrete residue type and side chain structure in a continuous latent space, on this dataset, we achieve new state-of-the-art results, with improvements of +54% in structural diversity and +27% in co-designability. To validate the broad utility of our approach, we further introduce Proteina Atomistica, a unified flow-based framework that jointly learns the distribution of protein backbone structure, discrete sequences, and atomistic side chains without latent variables. We again find that training on our new sequence-structure data dramatically boosts benchmark performance, improving \method's structural diversity by +73% and co-designability by +5%. Our work highlights the critical importance of aligned sequence-structure data for training high-performance de novo protein design models. Our new dataset https://catalog.ngc.nvidia.com/orgs/nvidia/teams/clara/resources/proteina-atomistica_data/files?version=release , the Consistency Distilled Synthetic Protein Database, is made available as an open-source resource.

q-bio.BM↗

arXiv2025

BioNeMo Framework: a modular, high-performance library for AI model development in drug discovery

Artificial Intelligence models encoding biology and chemistry are opening new routes to high-throughput and high-quality in-silico drug development. However, their training increasingly relies on computational scale, with recent protein language models (pLM) training on hundreds of graphical processing units (GPUs). We introduce the BioNeMo Framework to facilitate the training of computational biology and chemistry AI models across hundreds of GPUs. Its modular design allows the integration of individual components, such as data loaders, into existing workflows and is open to community contributions. We detail technical features of the BioNeMo Framework through use cases such as pLM pre-training and fine-tuning. On 256 NVIDIA A100s, BioNeMo Framework trains a three billion parameter BERT-based pLM on over one trillion tokens in 4.2 days. The BioNeMo Framework is open-source and free for everyone to use.

cs.LG↗

arXiv2025

Efficient Molecular Conformer Generation with SO(3)-Averaged Flow Matching and Reflow

Fast and accurate generation of molecular conformers is desired for downstream computational chemistry and drug discovery tasks. Currently, training and sampling state-of-the-art diffusion or flow-based models for conformer generation require significant computational resources. In this work, we build upon flow-matching and propose two mechanisms for accelerating training and inference of generative models for 3D molecular conformer generation. For fast training, we introduce the SO(3)-Averaged Flow training objective, which leads to faster convergence to better generation quality compared to conditional optimal transport flow or Kabsch-aligned flow. We demonstrate that models trained using SO(3)-Averaged Flow can reach state-of-the-art conformer generation quality. For fast inference, we show that the reflow and distillation methods of flow-based models enable few-steps or even one-step molecular conformer generation with high quality. The training techniques proposed in this work show a path towards highly efficient molecular conformer generation with flow-based models.

cs.LG↗

arXiv2025

ToxBench: A Binding Affinity Prediction Benchmark with AB-FEP-Calculated Labels for Human Estrogen Receptor Alpha

Protein-ligand binding affinity prediction is essential for drug discovery and toxicity assessment. While machine learning (ML) promises fast and accurate predictions, its progress is constrained by the availability of reliable data. In contrast, physics-based methods such as absolute binding free energy perturbation (AB-FEP) deliver high accuracy but are computationally prohibitive for high-throughput applications. To bridge this gap, we introduce ToxBench, the first large-scale AB-FEP dataset designed for ML development and focused on a single pharmaceutically critical target, Human Estrogen Receptor Alpha (ER$α$). ToxBench contains 8,770 ER$α$-ligand complex structures with binding free energies computed via AB-FEP with a subset validated against experimental affinities at 1.75 kcal/mol RMSE, along with non-overlapping ligand splits to assess model generalizability. Using ToxBench, we further benchmark state-of-the-art ML methods, and notably, our proposed DualBind model, which employs a dual-loss framework to effectively learn the binding energy function. The benchmark results demonstrate the superior performance of DualBind and the potential of ML to approximate AB-FEP at a fraction of the computational cost.

cs.LG↗

arXiv2025

Optimizing Data Distribution and Kernel Performance for Efficient Training of Chemistry Foundation Models: A Case Study with MACE

Chemistry Foundation Models (CFMs) that leverage Graph Neural Networks (GNNs) operating on 3D molecular graph structures are becoming indispensable tools for computational chemists and materials scientists. These models facilitate the understanding of matter and the discovery of new molecules and materials. In contrast to GNNs operating on a large homogeneous graphs, GNNs used by CFMs process a large number of geometric graphs of varying sizes, requiring different optimization strategies than those developed for large homogeneous GNNs. This paper presents optimizations for two critical phases of CFM training: data distribution and model training, targeting MACE - a state-of-the-art CFM. We address the challenge of load balancing in data distribution by formulating it as a multi-objective bin packing problem. We propose an iterative algorithm that provides a highly effective, fast, and practical solution, ensuring efficient data distribution. For the training phase, we identify symmetric tensor contraction as the key computational kernel in MACE and optimize this kernel to improve the overall performance. Our combined approach of balanced data distribution and kernel optimization significantly enhances the training process of MACE. Experimental results demonstrate a substantial speedup, reducing per-epoch execution time for training from 12 to 2 minutes on 740 GPUs with a 2.6M sample dataset.

cs.DC↗

arXiv2025

Proteina: Scaling Flow-based Protein Structure Generative Models

Recently, diffusion- and flow-based generative models of protein structures have emerged as a powerful tool for de novo protein design. Here, we develop Proteina, a new large-scale flow-based protein backbone generator that utilizes hierarchical fold class labels for conditioning and relies on a tailored scalable transformer architecture with up to 5x as many parameters as previous models. To meaningfully quantify performance, we introduce a new set of metrics that directly measure the distributional similarity of generated proteins with reference sets, complementing existing metrics. We further explore scaling training data to millions of synthetic protein structures and explore improved training and sampling recipes adapted to protein backbone generation. This includes fine-tuning strategies like LoRA for protein backbones, new guidance methods like classifier-free guidance and autoguidance for protein backbones, and new adjusted training objectives. Proteina achieves state-of-the-art performance on de novo protein backbone design and produces diverse and designable proteins at unprecedented length, up to 800 residues. The hierarchical conditioning offers novel control, enabling high-level secondary-structure guidance as well as low-level fold-specific generation.

cs.LG↗

arXiv2024

EquiJump: Protein Dynamics Simulation via SO(3)-Equivariant Stochastic Interpolants

Mapping the conformational dynamics of proteins is crucial for elucidating their functional mechanisms. While Molecular Dynamics (MD) simulation enables detailed time evolution of protein motion, its computational toll hinders its use in practice. To address this challenge, multiple deep learning models for reproducing and accelerating MD have been proposed drawing on transport-based generative methods. However, existing work focuses on generation through transport of samples from prior distributions, that can often be distant from the data manifold. The recently proposed framework of stochastic interpolants, instead, enables transport between arbitrary distribution endpoints. Building upon this work, we introduce EquiJump, a transferable SO(3)-equivariant model that bridges all-atom protein dynamics simulation time steps directly. Our approach unifies diverse sampling methods and is benchmarked against existing models on trajectory data of fast folding proteins. EquiJump achieves state-of-the-art results on dynamics simulation with a transferable model on all of the fast folding proteins.

cs.LG↗

arXiv2023

Domain-dependent surface adhesion in twisted few-layer graphene: Platform for moiré-assisted chemistry

Twisted van der Waals multilayers are widely regarded as a rich platform to access novel electronic phases, thanks to the multiple degrees of freedom such as layer thickness and twist angle that allow control of their electronic and chemical properties. Here, we propose that the stacking domains that form naturally due to the relative twist between successive layers act as an additional "knob" for controlling the behavior of these systems, and report the emergence and engineering of stacking domain-dependent surface chemistry in twisted few-layer graphene. Using mid-infrared near-field optical microscopy and atomic force microscopy, we observe a selective adhesion of metallic nanoparticles and liquid water at the domains with rhombohedral stacking configurations of minimally twisted double bi- and tri-layer graphene. Furthermore, we demonstrate that the manipulation of nanoparticles located at certain stacking domains can locally reconfigure the moiré superlattice in their vicinity at the μm-scale. In addition, we report first-principles simulations of the energetics of adhesion of metal atoms and water molecules on the stacking domains in an attempt to elucidate the origin of the observed selective adhesion. Our findings establish a new approach to controlling moiré-assisted chemistry and nanoengineering.

cond-mat.mtrl-sci↗

arXiv2020

A Systematic Approach to Generating Accurate Neural Network Potentials: the Case of Carbon

Availability of affordable and widely applicable interatomic potentials is the key needed to unlock the riches of modern materials modelling. Artificial neural network based approaches for generating potentials are promising; however neural network training requires large amounts of data, sampled adequately from an often unknown potential energy surface. Here we propose a self-consistent approach that is based on crystal structure prediction formalism and is guided by unsupervised data analysis, to construct an accurate, inexpensive and transferable artificial neural network potential. Using this approach, we construct an interatomic potential for Carbon and demonstrate its ability to reproduce first principles results on elastic and vibrational properties for diamond, graphite and graphene, as well as energy ordering and structural properties of a wide range of crystalline and amorphous phases.

cond-mat.mtrl-sci↗

arXiv2020

The CECAM Electronic Structure Library and the modular software development paradigm

First-principles electronic structure calculations are very widely used thanks to the many successful software packages available. Their traditional coding paradigm is monolithic, i.e., regardless of how modular its internal structure may be, the code is built independently from others, from the compiler up, with the exception of linear-algebra and message-passing libraries. This model has been quite successful for decades. The rapid progress in methodology, however, has resulted in an ever increasing complexity of those programs, which implies a growing amount of replication in coding and in the recurrent re-engineering needed to adapt to evolving hardware architecture. The Electronic Structure Library (\esl) was initiated by CECAM (European Centre for Atomic and Molecular Calculations) to catalyze a paradigm shift away from the monolithic model and promote modularization, with the ambition to extract common tasks from electronic structure programs and redesign them as free, open-source libraries. They include "heavy-duty" ones with a high degree of parallelisation, and potential for adaptation to novel hardware within them, thereby separating the sophisticated computer science aspects of performance optimization and re-engineering from the computational science done by scientists when implementing new ideas. It is a community effort, undertaken by developers of various successful codes, now facing the challenges arising in the new model. This modular paradigm will improve overall coding efficiency and enable specialists (computer scientists or computational scientists) to use their skills more effectively. It will lead to a more sustainable and dynamic evolution of software as well as lower barriers to entry for new developers.

cond-mat.mtrl-sci↗

arXiv2019

PANNA: Properties from Artificial Neural Network Architectures

Prediction of material properties from first principles is often a computationally expensive task. Recently, artificial neural networks and other machine learning approaches have been successfully employed to obtain accurate models at a low computational cost by leveraging existing example data. Here, we present a software package "Properties from Artificial Neural Network Architectures" (PANNA) that provides a comprehensive toolkit for creating neural network models for atomistic systems. Besides the core routines for neural network training, it includes data parser, descriptor builder and force-field generator suitable for integration within molecular dynamics packages. PANNA offers a variety of activation and cost functions, regularization methods, as well as the possibility of using fully-connected networks with custom size for each atomic species. PANNA benefits from the optimization and hardware-flexibility of the underlying TensorFlow engine which allows it to be used on multiple CPU/GPU/TPU systems, making it possible to develop and optimize neural network models based on large datasets.

physics.comp-ph↗

arXiv2016

Crystal Structure Prediction of Molecular Crystals from First Principles: Are we there yet?

Accurate molecular crystal structure prediction is a fundamental goal in academic and industrial condensed matter research and polymorphism is arguably the biggest obstacle on the way. We tackle this challenge in the difficult case of the repeatedly studied, abundantly used aminoacid Glycine that hosts still little-known phase transitions and we illustrate the current state of the field through this example. We demonstrate that the combination of recent progress in structure search algorithms with the latest advances in the description of van der Waals interactions in Density Functional Theory, supported by data-mining analysis, enables a leap in predictive power: we resolve, without prior empirical input, all known phases of glycine, as well as the structure of the previously unresolved $ζ$ phase after a decade of its experimental observation [Boldyreva et al. \textit{Z. Kristallogr.} \textbf{2005,} \textit{220,} 50-57]. The search for the well-established $α$ phase instead reveals the remaining challenges in exploring a polymorphic landscape.

cond-mat.mtrl-sci↗