Nature Machine Intelligence最新文献

{"title":"Learning the coupled dynamics of global climate modes","authors":"Yuan Yuan, Jingtao Ding, Zhongpu Qiu, Jingfang Fan, Yong Li","doi":"10.1038/s42256-026-01245-5","DOIUrl":"10.1038/s42256-026-01245-5","url":null,"abstract":"Global weather extremes, from monsoons to droughts, are shaped by a network of recurrent, coupled ocean–atmosphere patterns known as climate modes. These modes, spanning from the Pacific’s El Niño-Southern Oscillation to interconnected patterns in the Indian and Atlantic Oceans, form a dynamically linked global system governed by complex nonlinear interactions. Holistically forecasting this interconnected system—rather than treating modes in isolation or in simplified pairs as existing approaches do—remains a fundamental challenge in machine intelligence for complex systems. Here we introduce UniCM, a unified deep model for global climate modes forecasting. Its key innovation lies in a dual-branch architecture that learns the dynamics of a coupled system directly from data, achieving a truly unified prediction through the synergistic modelling of localized dynamics and their collective global couplings. UniCM achieves strong performance in unified global climate-mode forecasting, outperforming strong existing baselines and extending the skilful forecast lead time across multiple major climate modes. It successfully captured the diversity of historical events, from the extreme 1997–1998 El Niño to the prolonged and challenging 2020–2023 triple-dip La Niña. Beyond accuracy, UniCM offers interpretability; its internal attention mechanism identifies dynamic precursors and quantifies the structured inter-mode interactions that precede extreme climate events. Our results demonstrate that learning the coupled dynamics of climate modes as an interconnected system unlocks emergent predictability, laying a foundation for unified forecasting and data-driven insights that deepen our understanding of global ocean–atmosphere dynamics. Yuan et al. develop UniCM, a deep learning framework that models the coupled dynamics of global climate modes as an interconnected system, extending skilful forecasts for multiple climate modes and revealing cross-basin interactions.","PeriodicalId":48533,"journal":{"name":"Nature Machine Intelligence","volume":"8 6","pages":"930-941"},"PeriodicalIF":29.8,"publicationDate":"2026-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148300159","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Human–AI interactions reshape the self and our social networks","authors":"Andreia Sofia Teixeira, Sukhwinder Singh Shergill, Guy Laban","doi":"10.1038/s42256-026-01248-2","DOIUrl":"10.1038/s42256-026-01248-2","url":null,"abstract":"","PeriodicalId":48533,"journal":{"name":"Nature Machine Intelligence","volume":"8 6","pages":"858-859"},"PeriodicalIF":29.8,"publicationDate":"2026-05-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148300151","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Generalizable mutation-effect prediction across adaptive immune recognition via unified multimodal framework","authors":"Rong Han, Yumeng Zhang, Xiaohong Liu, Lei Fu, Tong Pan, Jing Xu, Xiaoyu Wang, Peidong Zhang, Xuanzhong Chen, Jiesi Lei, Wuyang Lan, Changwei Ji, Shuguang Cui, Song Wu, Jiangning Song, Ting Chen, Guangyu Wang","doi":"10.1038/s42256-026-01243-7","DOIUrl":"10.1038/s42256-026-01243-7","url":null,"abstract":"Adaptive immunity is a central defence system essential for long-term and highly specific protection against pathogens through the precise molecular recognition of antigens by lymphocytes. However, predicting how mutations reshape these interactions remains a major challenge. Although previous computational approaches leverage large-scale pretraining for mutation-effect predictions, most are designed for specific tasks or modalities and struggle to generalize across the heterogeneous, multimodal landscape of immune recognition. Here we introduce UniAIR, a modular, multimodal framework for the accurate and generalizable prediction of mutation effects across immune recognition scenarios. UniAIR integrates a standardized data pipeline, an interface-centric sequence–structure fusion transformer that integrates evolutionary information with geometric representations, and a suite of extensions for multiexpert consensus and adaptation to predicted structure inputs. We comprehensively evaluated UniAIR through large-scale benchmarking and independent tests across immunological tasks. The evaluation covered both extracellular and intracellular immune recognition, including antibody maturation, antigen escape, TCR–pHLA optimization and analyses in which experimental structures were unavailable. Extensive experiments show that UniAIR achieves state-of-the-art performance and delivers robust predictions with minimal task-specific tuning. In particular, UniAIR successfully performed multiround peptide optimization of a TCR–pHLA complex under sparse feedback and identified key functional mutations in incomplete antibody–antigen structures. Together, UniAIR establishes a unified computational foundation for mapping mutation landscapes, advancing understanding of adaptive immune recognition and accelerating immunotherapeutic design. Han et al. introduce UniAIR, a unified AI framework that predicts mutation effects across diverse immune recognition settings. The approach enables more generalizable modelling of antibody, antigen and T cell receptor interactions.","PeriodicalId":48533,"journal":{"name":"Nature Machine Intelligence","volume":"8 6","pages":"913-929"},"PeriodicalIF":29.8,"publicationDate":"2026-05-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148300199","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A large-scale unified deep learning model for peptide mass spectrum interpretation trained on multimodal data","authors":"Jiale Zhao, Pengzhi Mao, Kaifei Wang, Yiming Li, Yaping Peng, Ranfei Chen, Shuqi Lu, Xiaohong Ji, Jiaxiang Ding, Xin Zhang, Yucheng Liao, Weinan E, Han Wen, Weijie Zhang, Hao Chi","doi":"10.1038/s42256-026-01234-8","DOIUrl":"10.1038/s42256-026-01234-8","url":null,"abstract":"Deep learning has advanced mass spectrometry data interpretation, yet most models remain feature extractors rather than unified scoring frameworks. We present pUniFind, a large-scale multimodal foundational model in proteomics that integrates open end-to-end peptide-spectrum scoring with open, zero-shot de novo sequencing. Trained on over 100 million open search-derived spectra, pUniFind aligns spectral and peptide modalities through cross-modality prediction alongside other carefully designed pretraining tasks. Consequently, further benefiting from its open scoring capability, pUniFind outperforms traditional engines across diverse datasets, notably achieving a 42.6% increase in identified peptides in immunopeptidomics. We propose two de novo sequencing workflows to support different applications. For modification-rich de novo sequencing, pUniFind identifies 60% more peptide–spectrum matches than existing de novo methods despite a 300 times larger search space. For regular de novo sequencing, pUniFind recovers an additional 38.5% of peptides, including 1,891 that map to the genome but are absent from reference proteomes. Crucially, it achieves this while preserving full fragment ion coverage and maintaining high consistency with database-search-based methods. Furthermore, a quality control module based on deep learning-derived features increases the consistency of results with RNA-Seq evidence from 65.4% to 85.0%. These results establish a unified, scalable deep learning framework for proteomic analysis, offering improved sensitivity, modification coverage, and interpretability. pUniFind is a large-scale deep learning model for proteomics that unifies peptide–spectrum scoring and open de novo sequencing. Trained on over 100 million spectra, it substantially improves peptide identification and modification discovery.","PeriodicalId":48533,"journal":{"name":"Nature Machine Intelligence","volume":"8 6","pages":"997-1011"},"PeriodicalIF":29.8,"publicationDate":"2026-05-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148300158","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Neural operators for free-boundary problems","authors":"Constantinos Siettos","doi":"10.1038/s42256-026-01238-4","DOIUrl":"10.1038/s42256-026-01238-4","url":null,"abstract":"Free boundary problems, such as modelling glacier melt, are difficult to capture with neural operators. A new framework addresses this challenge by leveraging the mathematical principle of topological conjugacy.","PeriodicalId":48533,"journal":{"name":"Nature Machine Intelligence","volume":"8 5","pages":"647-648"},"PeriodicalIF":23.9,"publicationDate":"2026-05-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148012349","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Deep neural operator for free boundary problems","authors":"Zongjia Long, Qi Zhou, Aiqing Zhu, Dong Dai, Yiqun Liu","doi":"10.1038/s42256-026-01233-9","DOIUrl":"10.1038/s42256-026-01233-9","url":null,"abstract":"Free boundary problems, characterized by partial differential equations defined on a priori unknown domains, arise across diverse scientific and engineering disciplines, from quantum physics to biomedical applications. Traditional numerical approaches face crucial limitations. Although recent advances in neural operators have revolutionized the solving of partial differential equations by learning mappings between function spaces, existing frameworks remain constrained to predefined domains, rendering them inapplicable to free boundary problems. Here we introduce the free boundary neural operator, a universal framework that overcomes this fundamental limitation by leveraging topological conjugacy between dynamical systems. The free boundary neural operator approximates both the flow map of a conjugate system and the homeomorphism linking it to the original free boundary problem, which enables predictions on evolving domains without prior geometric knowledge. Crucially, we provide an approximation theorem guaranteeing the theoretical feasibility of the method. We have comprehensively demonstrated the efficacy of the free boundary neural operator in numerical experiments spanning phase transitions, non-convex geometries and multi-physics systems. The work marks a turning point in free boundary simulations, as the use of neural networks unlocks both speed and precision. Long et al. introduce a neural operator method to solve free boundary problems with high precision. The framework shows promise for real-time predictions in clinical applications, particularly in simulating tumour growth.","PeriodicalId":48533,"journal":{"name":"Nature Machine Intelligence","volume":"8 5","pages":"806-817"},"PeriodicalIF":23.9,"publicationDate":"2026-05-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.nature.comhttps://www.nature.com/articles/s42256-026-01233-9.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148012350","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Stop ‘tokenmaxxing’ and deploy AI sensibly instead","authors":"","doi":"10.1038/s42256-026-01253-5","DOIUrl":"10.1038/s42256-026-01253-5","url":null,"abstract":"Companies, tech workers and researchers are in a frenzy to embed agentic AI into their workflows, locked in a self-imposed race not to fall behind. There must be a better way to make use of AI technology.","PeriodicalId":48533,"journal":{"name":"Nature Machine Intelligence","volume":"8 5","pages":"641-641"},"PeriodicalIF":23.9,"publicationDate":"2026-05-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.nature.comhttps://www.nature.com/articles/s42256-026-01253-5.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148012345","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Plagiarism of ideas in the age of generative artificial intelligence","authors":"David Resnik, Mohammad Hosseini","doi":"10.1038/s42256-026-01247-3","DOIUrl":"10.1038/s42256-026-01247-3","url":null,"abstract":"Generative artificial intelligence (GenAI) tools are challenging our understanding of plagiarism. How should we deal with plagiarism of ideas if this misbehaviour is increasingly common, and it is extremely difficult to prove when GenAI is involved? Definitions of research misconduct that specifically address the use of GenAI tools are needed.","PeriodicalId":48533,"journal":{"name":"Nature Machine Intelligence","volume":"8 6","pages":"860-861"},"PeriodicalIF":29.8,"publicationDate":"2026-05-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148151176","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"SpecGP as a transformer-based model for predicting energy-adaptable structural spectra of glycopeptides","authors":"Xianyong Wang, Rui Song, Zhuangzhuang Feng, Shisheng Sun","doi":"10.1038/s42256-026-01246-4","DOIUrl":"10.1038/s42256-026-01246-4","url":null,"abstract":"Deep learning has shown great promise for high-throughput spectral library construction in glycoproteomics. Nevertheless, precise prediction of the structural spectra of intact N-glycopeptides remains challenging due to their complex and high-dimensional information. Here we present SpecGP, a transformer-based architecture featuring an attention-enhanced glycan fragment encoding strategy with multilayer perceptrons for accurate glycopeptide spectrum prediction. The model improves spectral differentiation among glycopeptides by expanding fragment ion coverage while retaining high prediction accuracy. By predicting mass spectra at multiple collision energies, SpecGP maximizes the detection of key diagnostic ions and ensures broad compatibility with diverse experimental datasets. It also improves retention time prediction by using a dual-task framework. In applications, SpecGP effectively enhances isomeric discrimination by incorporating a self-supervised weighting training strategy and boosts glycopeptide identification via rescoring. The capability of glycan structure discrimination can be further strengthened and validated by complementary diagnostic ions with dynamic intensities in multi-energy spectra. SpecGP enhances fragment ion coverage to enable the prediction of N-glycopeptide structural spectra across diverse collision energies, thereby improving isomer discrimination and boosting identification confidence through rescoring.","PeriodicalId":48533,"journal":{"name":"Nature Machine Intelligence","volume":"8 6","pages":"942-954"},"PeriodicalIF":29.8,"publicationDate":"2026-05-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148300157","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Immunotherapy drug target identification using machine learning and patient-derived tumour explant validation","authors":"Marcellus Augustine, Nuno Rocha Nene, Hongchang Fu, Christopher L. Pinder, Lorena Ligammari, Alexander P. Simpson, Irene Sanz-Fernández, Krupa Thakkar, Danwen Qian, Evelyn Fitzsimons, Benjamin S. Simpson, Roberto Vendramin, Andrea Castro, Heather Niederer, Samra Turajlic, Sergio A. Quezada, Nicholas McGranahan, Chris Watkins, Charles Swanton, Kevin Litchfield","doi":"10.1038/s42256-026-01201-3","DOIUrl":"10.1038/s42256-026-01201-3","url":null,"abstract":"Immunotherapy has revolutionized cancer treatment, yet only a minority of individuals respond clinically, necessitating alternative strategies that can benefit these patients. Novel immuno-oncology targets may achieve this through bypassing resistance mechanisms to standard therapies. We introduce Mining Immunotherapy Drug tArgetS (MIDAS), a multimodal graph neural network system for immuno-oncology target discovery. MIDAS leverages gene interactions, multi-omic patient profiles, immune cell biology, antigen processing, disease associations and phenotypic consequences of genetic perturbations. It generalizes to time-sliced data, outcompetes state-of-the-art baselines (including OpenTargets) and ranks approved targets above those in clinical development. Moreover, MIDAS recovers immunotherapy-response-associated genes in unseen patients, thereby capturing immunotherapy response determinants. Interpretability analyses reveal a reliance on autoimmunity, regulatory networks and immuno-oncology pathways. Functionally perturbing oncostatin M–oncostatin M receptor signalling, a proposed MIDAS target, in TRACERx melanoma-patient-derived explants yielded reduced dysfunctional CD8+ T cells, which associate with immunotherapy response, and reduced CCL4 levels. Furthermore, oncostatin M and oncostatin M receptor expression is associated with altered T cell and macrophage profiles in bulk transcriptomic data from patient samples. These data are consistent with a role for oncostatin M–oncostatin M in modulating the tumour microenvironment towards immunosuppressive, tumour-promoting phenotypes. Our results present a machine learning framework for analysing multimodal data for immuno-oncology target discovery. Augustine et al. present a multimodal graph neural network that identifies cancer immunotherapy targets. It distinguishes approved and prospective targets, and promising candidates are validated using a clinically relevant patient-derived platform.","PeriodicalId":48533,"journal":{"name":"Nature Machine Intelligence","volume":"8 5","pages":"670-689"},"PeriodicalIF":23.9,"publicationDate":"2026-05-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.nature.comhttps://www.nature.com/articles/s42256-026-01201-3.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148012342","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}