Nature computational science最新文献

Hybrid modeling for improved structure prediction of large protein complexes. 改进大蛋白质复合物结构预测的混合模型。

IF 20.3

Nature computational science Pub Date : 2026-10-15 DOI: 10.1038/s43588-026-01067-w

引用次数: 0

HDOCK-Multimer: improved structure prediction of large protein complexes by integrating docking and combinatorial assembly. hdock - multitimer：通过整合对接和组合组装，改进了大蛋白复合物的结构预测。

IF 20.3

Nature computational science Pub Date : 2026-10-14 DOI: 10.1038/s43588-026-01062-1

Xuan Yao, Yifan Ya, Hao Li, Sheng-You Huang

{"title":"HDOCK-Multimer: improved structure prediction of large protein complexes by integrating docking and combinatorial assembly.","authors":"Xuan Yao, Yifan Ya, Hao Li, Sheng-You Huang","doi":"10.1038/s43588-026-01062-1","DOIUrl":"10.1038/s43588-026-01062-1","url":null,"abstract":"Accurate prediction of large protein assembly structures is crucial for understanding biological processes, yet remains a challenge for end-to-end approaches such as AlphaFold owing to their large size and intricate multi-chain interactions. Docking methods can handle large proteins, but suffer from the huge combinatorial binding space of multiple chains. Assembly methods offer an alternative way, but critically rely on the quality of predicted subcomponents. Here we present HDOCK-Multimer (HDM), a framework that integrates docking and combinatorial assembly for structure prediction of large protein complexes. HDM efficiently reduces reliance on subcomponent quality through docking, while leveraging the pairwise interactions of subcomponents through assembly. When evaluated on 3 benchmarks of 35 large heteromeric complexes, 172 large protein complexes and 7 CASP15 targets, HDM substantially outperforms state-of-the-art methods including CombFold, MoLPC, AlphaFold-Multimer and AlphaFold3. In addition, HDM also demonstrates the capacity to predict the stoichiometry and model the complex without stoichiometry input.","PeriodicalId":74246,"journal":{"name":"Nature computational science","volume":" ","pages":""},"PeriodicalIF":20.3,"publicationDate":"2026-10-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149185998","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

LLMs predict material properties based on atom types and not atom positions. llm基于原子类型而不是原子位置来预测材料的性质。

IF 20.3

Nature computational science Pub Date : 2026-10-13 DOI: 10.1038/s43588-026-01044-3

引用次数: 0

HIP: Hessian Interatomic Potentials without derivatives. 无导数的黑森原子相互作用势。

IF 20.3

Nature computational science Pub Date : 2026-10-13 DOI: 10.1038/s43588-026-01066-x

Andreas Burger, Luca Thiede, Nikolaj Rønne, Nandita Vijaykumar, Varinia Bernales, Tejs Vegge, Arghya Bhowmik, Alán Aspuru-Guzik

{"title":"HIP: Hessian Interatomic Potentials without derivatives.","authors":"Andreas Burger, Luca Thiede, Nikolaj Rønne, Nandita Vijaykumar, Varinia Bernales, Tejs Vegge, Arghya Bhowmik, Alán Aspuru-Guzik","doi":"10.1038/s43588-026-01066-x","DOIUrl":"https://doi.org/10.1038/s43588-026-01066-x","url":null,"abstract":"Molecular Hessians, the second derivatives of the potential energy, are fundamental to many workflows in computational chemistry. Usually, accurate Hessians are computationally expensive to calculate and scale poorly with system size, whether computed using quantum chemistry methods or machine learning interatomic potentials. Here we introduce Hessian Interatomic Potentials (HIPs), a deep learning model that directly predicts Hessians without relying on automatic differentiation or finite differences. To do so, we construct SE(3)-equivariant, symmetric Hessians from irreducible representation (irrep) features up to degree l = 2, computed by a graph neural network. HIP Hessians are one to two orders of magnitude faster, more accurate, more memory efficient, easier to train and exhibit favorable scaling with system size. We validate our predictions across a wide range of downstream tasks, demonstrating consistently superior performance in transition state search, geometry optimization, zero-point energy corrections and vibrational analysis. We open source the HIP code and model weights.","PeriodicalId":74246,"journal":{"name":"Nature computational science","volume":" ","pages":""},"PeriodicalIF":20.3,"publicationDate":"2026-10-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149186122","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Less can be more for predicting properties with large language models. 对于使用大型语言模型预测属性来说，少即是多。

IF 20.3

Nature computational science Pub Date : 2026-10-12 DOI: 10.1038/s43588-026-01045-2

Nawaf Alampara, Santiago Miret, Kevin Maik Jablonka

{"title":"Less can be more for predicting properties with large language models.","authors":"Nawaf Alampara, Santiago Miret, Kevin Maik Jablonka","doi":"10.1038/s43588-026-01045-2","DOIUrl":"10.1038/s43588-026-01045-2","url":null,"abstract":"Predicting properties from coordinate-category data-sets of vectors paired with categorical information-is fundamental to computational science, manifesting in materials science as predicting formation energies or elastic moduli from crystal structures. While large language models (LLMs) have increasingly been applied to such tasks, optimal strategies for achieving reliable predictions remain elusive. Here we report fundamental limitations in LLM's ability to learn from coordinate information in coordinate-category data. Through systematic experiments using synthetic datasets with tunable coordinate and category contributions, and a comprehensive benchmarking framework (MatText), we find that LLMs consistently fail to capture coordinate information while excelling at category patterns. This geometric blindness persists regardless of model size (up to 70 billion parameters), dataset scale (up to 2 × 106 structures) or text representation strategy. For material property prediction tasks dominated by structural effects, specialized geometric architectures consistently outperform LLMs, as evidenced by a clear 'Graph Neural Network Language Model (GNN-LM) wall' in performance benchmarks. Our analysis provides concrete guidelines for architecture selection in scientific machine learning.","PeriodicalId":74246,"journal":{"name":"Nature computational science","volume":" ","pages":""},"PeriodicalIF":20.3,"publicationDate":"2026-10-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149168247","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Epidemiological modeling in the age of artificial intelligence. 人工智能时代的流行病学建模。

IF 20.3

Nature computational science Pub Date : 2026-10-09 DOI: 10.1038/s43588-026-01070-1

Max S Y Lau, Matthew J Ferrari, Wei Jin

引用次数: 0

Steering complex rhythms. 驾驭复杂的节奏。

IF 20.3

Nature computational science Pub Date : 2026-10-09 DOI: 10.1038/s43588-026-01064-z

Fan Jin

引用次数: 0

Photonic circuits become trainable AI hardware. 光子电路成为可训练的AI硬件。

IF 20.3

Nature computational science Pub Date : 2026-10-09 DOI: 10.1038/s43588-026-01071-0

引用次数: 0

Advancing chemical reaction prediction in data-scarce drug discovery with active and geometric deep learning. 利用主动和几何深度学习推进数据稀缺药物发现中的化学反应预测。

IF 20.3

Nature computational science Pub Date : 2026-10-09 DOI: 10.1038/s43588-026-01063-0

Mason Minot, Yannick Stenzhorn, Jens Wolfard, Sebastian Strobel, Philippe Jablonski, Daniel Zimmerli, Martin Binder, Uwe Grether, Rainer E Martin, Alex T Müller, David F Nippa, Kenneth Atz, Gisbert Schneider

{"title":"Advancing chemical reaction prediction in data-scarce drug discovery with active and geometric deep learning.","authors":"Mason Minot, Yannick Stenzhorn, Jens Wolfard, Sebastian Strobel, Philippe Jablonski, Daniel Zimmerli, Martin Binder, Uwe Grether, Rainer E Martin, Alex T Müller, David F Nippa, Kenneth Atz, Gisbert Schneider","doi":"10.1038/s43588-026-01063-0","DOIUrl":"https://doi.org/10.1038/s43588-026-01063-0","url":null,"abstract":"The success of generative molecular design depends on the synthetic accessibility of the molecules it creates. While computational models can help predict chemical reactions, their current effectiveness is limited by a lack of reference data and the difficulty of accurately predicting reaction regioselectivity. Here, to address this, we present an integrated closed-loop workflow that combines efficient reaction data acquisition with high-accuracy deep learning. We developed an active learning strategy, driven by a tree-based ensemble, to guide the acquisition of the most informative laboratory experiments. This experimental campaign yielded a diverse benchmark for C-H borylation. The bespoke dataset allowed us to train and evaluate geometric graph neural networks. We demonstrate that augmenting these symmetry-aware models with self-supervised auxiliary tasks consistently improves performance for both reaction outcome prediction and regioselectivity prediction. Prospective tests on unseen substrates, featuring challenging N-heteroaryl motifs, pinpointed the correct borylation positions in all cases, illustrating the practical utility of our workflow for efficient drug optimization.","PeriodicalId":74246,"journal":{"name":"Nature computational science","volume":" ","pages":""},"PeriodicalIF":20.3,"publicationDate":"2026-10-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149148820","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Photonic neuromorphic learning via generalized in situ physical gradient descent. 通过广义原位物理梯度下降的光子神经形态学习。

IF 20.3

Nature computational science Pub Date : 2026-10-05 DOI: 10.1038/s43588-026-01057-y

Tiankuang Zhou, Yun Zhao, Shanglong Li, Guocheng Shao, Ruqi Huang, Lu Fang

引用次数: 0