Pub Date : 2024-10-01DOI: 10.1038/s42256-024-00894-8
Enrica Tricomi, Francesco Missiroli, Michele Xiloyannis, Nicola Lotti, Xiaohui Zhang, Marios Stefanakis, Maximilian Theisen, Jürgen Bauer, Clemens Becker, Lorenzo Masia
Peoples'' walking efficiency declines as they grow older, posing constraints on mobility, and affecting independence and quality of life. Although wearable assistive technologies are recognized as a potential solution for age-related movement challenges, few have proven effective for older adults, predominantly within controlled laboratory experiments. Here we present WalkON, a pair of soft robotic shorts designed to enhance walking efficiency for older individuals by assisting hip flexion. The system features a compact and lightweight tendon-driven design, using a controller based on natural leg movements to autonomously assist leg propagation. To assess WalkON''s impact on daily walking, we initially conducted a technology assessment with young adults on a demanding outdoor uphill 500 m hiking trail. We then validated our findings with a group of older adults walking on a flat outdoor 400 m track. WalkON considerably reduced the metabolic cost of transport by 17.79% for young adults during uphill walking. At the same time, participants reported high perceived control over their voluntary movements (a self-reported mean score of 6.20 out of 7 on a Likert scale). Similarly, older adults reduced their metabolic cost by 10.48% when using WalkON during level ground walking, while retaining a strong sense of movement control (mean score of 6.09 out of 7). These findings emphasize the potential of wearable assistive devices to improve efficiency in outdoor walking, suggesting promising implications for promoting physical well-being and advancing mobility, particularly during the later stages of life. Walking efficiency declines in older adults. To address this challenge, Tricomi and colleagues present a pair of lightweight, soft robotic shorts that enhance walking efficiency for older adults by assisting leg mobility. This method improves energy efficiency on outdoor tracks while maintaining the users’ natural movement control.
{"title":"Soft robotic shorts improve outdoor walking efficiency in older adults","authors":"Enrica Tricomi, Francesco Missiroli, Michele Xiloyannis, Nicola Lotti, Xiaohui Zhang, Marios Stefanakis, Maximilian Theisen, Jürgen Bauer, Clemens Becker, Lorenzo Masia","doi":"10.1038/s42256-024-00894-8","DOIUrl":"10.1038/s42256-024-00894-8","url":null,"abstract":"Peoples'' walking efficiency declines as they grow older, posing constraints on mobility, and affecting independence and quality of life. Although wearable assistive technologies are recognized as a potential solution for age-related movement challenges, few have proven effective for older adults, predominantly within controlled laboratory experiments. Here we present WalkON, a pair of soft robotic shorts designed to enhance walking efficiency for older individuals by assisting hip flexion. The system features a compact and lightweight tendon-driven design, using a controller based on natural leg movements to autonomously assist leg propagation. To assess WalkON''s impact on daily walking, we initially conducted a technology assessment with young adults on a demanding outdoor uphill 500 m hiking trail. We then validated our findings with a group of older adults walking on a flat outdoor 400 m track. WalkON considerably reduced the metabolic cost of transport by 17.79% for young adults during uphill walking. At the same time, participants reported high perceived control over their voluntary movements (a self-reported mean score of 6.20 out of 7 on a Likert scale). Similarly, older adults reduced their metabolic cost by 10.48% when using WalkON during level ground walking, while retaining a strong sense of movement control (mean score of 6.09 out of 7). These findings emphasize the potential of wearable assistive devices to improve efficiency in outdoor walking, suggesting promising implications for promoting physical well-being and advancing mobility, particularly during the later stages of life. Walking efficiency declines in older adults. To address this challenge, Tricomi and colleagues present a pair of lightweight, soft robotic shorts that enhance walking efficiency for older adults by assisting leg mobility. This method improves energy efficiency on outdoor tracks while maintaining the users’ natural movement control.","PeriodicalId":48533,"journal":{"name":"Nature Machine Intelligence","volume":"6 10","pages":"1145-1155"},"PeriodicalIF":18.8,"publicationDate":"2024-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.nature.com/articles/s42256-024-00894-8.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142360330","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}
Pub Date : 2024-09-27DOI: 10.1038/s42256-024-00901-y
Ziyan Feng, Jingyang Chen, Youlong Hai, Xuelian Pang, Kun Zheng, Chenglong Xie, Xiujuan Zhang, Shengqing Li, Chengjuan Zhang, Kangdong Liu, Lili Zhu, Xiaoyong Hu, Shiliang Li, Jie Zhang, Kai Zhang, Honglin Li
Neoantigens are promising targets for immunotherapy by eliciting immune response and removing cancer cells with high specificity, low toxicity and ease of personalization. However, identifying effective neoantigens remains difficult because of the complex interactions among T cell receptors, antigens and human leucocyte antigen sequences. In this study, we integrate important physical and biological priors with the Transformer model and propose the physics-inspired sliding transformer (PISTE). In PISTE, the conventional, data-driven attention mechanism is replaced with physics-driven dynamics that steers the positioning of amino acid residues along the gradient field of their interactions. This allows navigating the intricate landscape of biosequence interactions intelligently, leading to improved accuracy in T cell receptor–antigen–human leucocyte antigen binding prediction and robust generalization to rare sequences. Furthermore, PISTE effectively recovers residue-level contact relationships even in the absence of three-dimensional structure training data. We applied PISTE in a multitude of immunogenic tumour types to pinpoint neoantigens and discern neoantigen-reactive T cells. In a prospective study of prostate cancer, 75% of the patients elicited immune responses through PISTE-predicted neoantigens. Predicting TCR–antigen–human leucocyte antigen binding opens the door to neoantigen identification. In this study, a physics-inspired sliding transformer (PISTE) system is used to guide the positioning of amino acid residues along the gradient field of their interactions, boosting binding prediction accuracy.
新抗原具有高特异性、低毒性和易于个性化等特点,可诱发免疫反应并清除癌细胞,是很有前途的免疫疗法靶点。然而,由于 T 细胞受体、抗原和人类白细胞抗原序列之间存在复杂的相互作用,识别有效的新抗原仍然十分困难。在这项研究中,我们将重要的物理和生物先验与变压器模型相结合,提出了物理启发滑动变压器(PISTE)。在 PISTE 中,传统的数据驱动注意力机制被物理驱动动力学所取代,物理驱动动力学引导氨基酸残基沿着其相互作用的梯度场进行定位。这样就能智能地浏览错综复杂的生物序列相互作用,从而提高 T 细胞受体-抗原-人类白细胞抗原结合预测的准确性,并对罕见序列进行稳健的泛化。此外,即使在没有三维结构训练数据的情况下,PISTE 也能有效地恢复残基级接触关系。我们将 PISTE 应用于多种免疫原性肿瘤类型,以确定新抗原并识别新抗原反应 T 细胞。在一项前瞻性前列腺癌研究中,75% 的患者通过 PISTE 预测的新抗原产生了免疫反应。
{"title":"Sliding-attention transformer neural architecture for predicting T cell receptor–antigen–human leucocyte antigen binding","authors":"Ziyan Feng, Jingyang Chen, Youlong Hai, Xuelian Pang, Kun Zheng, Chenglong Xie, Xiujuan Zhang, Shengqing Li, Chengjuan Zhang, Kangdong Liu, Lili Zhu, Xiaoyong Hu, Shiliang Li, Jie Zhang, Kai Zhang, Honglin Li","doi":"10.1038/s42256-024-00901-y","DOIUrl":"10.1038/s42256-024-00901-y","url":null,"abstract":"Neoantigens are promising targets for immunotherapy by eliciting immune response and removing cancer cells with high specificity, low toxicity and ease of personalization. However, identifying effective neoantigens remains difficult because of the complex interactions among T cell receptors, antigens and human leucocyte antigen sequences. In this study, we integrate important physical and biological priors with the Transformer model and propose the physics-inspired sliding transformer (PISTE). In PISTE, the conventional, data-driven attention mechanism is replaced with physics-driven dynamics that steers the positioning of amino acid residues along the gradient field of their interactions. This allows navigating the intricate landscape of biosequence interactions intelligently, leading to improved accuracy in T cell receptor–antigen–human leucocyte antigen binding prediction and robust generalization to rare sequences. Furthermore, PISTE effectively recovers residue-level contact relationships even in the absence of three-dimensional structure training data. We applied PISTE in a multitude of immunogenic tumour types to pinpoint neoantigens and discern neoantigen-reactive T cells. In a prospective study of prostate cancer, 75% of the patients elicited immune responses through PISTE-predicted neoantigens. Predicting TCR–antigen–human leucocyte antigen binding opens the door to neoantigen identification. In this study, a physics-inspired sliding transformer (PISTE) system is used to guide the positioning of amino acid residues along the gradient field of their interactions, boosting binding prediction accuracy.","PeriodicalId":48533,"journal":{"name":"Nature Machine Intelligence","volume":"6 10","pages":"1216-1230"},"PeriodicalIF":18.8,"publicationDate":"2024-09-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.nature.com/articles/s42256-024-00901-y.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142325411","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}
Pub Date : 2024-09-27DOI: 10.1038/s42256-024-00909-4
Janne Heirman, Wout Bittremieux
We present an in-depth exploration of the Metabolite Inference with Spectrum Transformers (MIST) tool for annotating small-molecule mass spectrometry (MS) data, focusing on its reproducibility and generalizability. MIST innovates by integrating a ‘chemical formula transformer’ to process tandem MS spectra, aiming to bridge the substantial knowledge gap in untargeted MS studies, in which only a fraction of spectra are confidently annotated. Here we critically assessed the reproducibility of MIST by following the tool’s original training and testing protocols, encountering minor challenges but largely succeeding in replicating the results. We also evaluated the generalizability of MIST by applying it to an external dataset from the Critical Assessment of Small Molecule Identification 2022 challenge, showing insights into the model’s performance on previously unseen data. An ablation study further investigated the impact of various model features on database retrieval performance, suggesting that some algorithmic complexities may not significantly enhance the performance. Through rigorous evaluation, this study underscores the challenges and considerations in developing robust computational tools for MS data analysis. We advocate community-wide efforts in benchmarking, transparency and data sharing to foster advancements in metabolomics and computational biology.
{"title":"Reusability report: Annotating metabolite mass spectra with domain-inspired chemical formula transformers","authors":"Janne Heirman, Wout Bittremieux","doi":"10.1038/s42256-024-00909-4","DOIUrl":"https://doi.org/10.1038/s42256-024-00909-4","url":null,"abstract":"<p>We present an in-depth exploration of the Metabolite Inference with Spectrum Transformers (MIST) tool for annotating small-molecule mass spectrometry (MS) data, focusing on its reproducibility and generalizability. MIST innovates by integrating a ‘chemical formula transformer’ to process tandem MS spectra, aiming to bridge the substantial knowledge gap in untargeted MS studies, in which only a fraction of spectra are confidently annotated. Here we critically assessed the reproducibility of MIST by following the tool’s original training and testing protocols, encountering minor challenges but largely succeeding in replicating the results. We also evaluated the generalizability of MIST by applying it to an external dataset from the Critical Assessment of Small Molecule Identification 2022 challenge, showing insights into the model’s performance on previously unseen data. An ablation study further investigated the impact of various model features on database retrieval performance, suggesting that some algorithmic complexities may not significantly enhance the performance. Through rigorous evaluation, this study underscores the challenges and considerations in developing robust computational tools for MS data analysis. We advocate community-wide efforts in benchmarking, transparency and data sharing to foster advancements in metabolomics and computational biology.</p>","PeriodicalId":48533,"journal":{"name":"Nature Machine Intelligence","volume":"3 1","pages":""},"PeriodicalIF":23.8,"publicationDate":"2024-09-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142325410","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}
Pub Date : 2024-09-27DOI: 10.1038/s42256-024-00895-7
Alexander Rodríguez, Harshavardhan Kamarthi, Pulak Agarwal, Javen Ho, Mira Patel, Suchet Sapre, B. Aditya Prakash
The COVID-19 pandemic emphasized the importance of epidemic forecasting for decision makers in multiple domains, ranging from public health to the economy. Forecasting epidemic progression is a non-trivial task due to multiple confounding factors, such as human behaviour, pathogen dynamics and environmental conditions. However, the surge in research interest and initiatives from public health and funding agencies has fuelled the availability of new data sources that capture previously unobservable aspects of disease spread, paving the way for a spate of ‘data-centred’ computational solutions that show promise for enhancing our forecasting capabilities. Here we discuss various methodological and practical advances and introduce a conceptual framework to navigate through them. First we list relevant datasets, such as symptomatic online surveys, retail and commerce, mobility and genomics data. Next we consider methods, focusing on recent data-driven statistical and deep learning-based methods, as well as hybrid models that combine domain knowledge of mechanistic models with the flexibility of statistical approaches. We also discuss experiences and challenges that arise in the real-world deployment of these forecasting systems, including decision-making informed by forecasts. Finally, we highlight some challenges and open problems found across the forecasting pipeline to enable robust future pandemic preparedness. Forecasting epidemic progression is a complex task influenced by various factors, including human behaviour, pathogen dynamics and environmental conditions. Rodríguez, Kamarthi and colleagues provide a review of machine learning methods for epidemic forecasting from a data-centric computational perspective.
{"title":"Machine learning for data-centric epidemic forecasting","authors":"Alexander Rodríguez, Harshavardhan Kamarthi, Pulak Agarwal, Javen Ho, Mira Patel, Suchet Sapre, B. Aditya Prakash","doi":"10.1038/s42256-024-00895-7","DOIUrl":"10.1038/s42256-024-00895-7","url":null,"abstract":"The COVID-19 pandemic emphasized the importance of epidemic forecasting for decision makers in multiple domains, ranging from public health to the economy. Forecasting epidemic progression is a non-trivial task due to multiple confounding factors, such as human behaviour, pathogen dynamics and environmental conditions. However, the surge in research interest and initiatives from public health and funding agencies has fuelled the availability of new data sources that capture previously unobservable aspects of disease spread, paving the way for a spate of ‘data-centred’ computational solutions that show promise for enhancing our forecasting capabilities. Here we discuss various methodological and practical advances and introduce a conceptual framework to navigate through them. First we list relevant datasets, such as symptomatic online surveys, retail and commerce, mobility and genomics data. Next we consider methods, focusing on recent data-driven statistical and deep learning-based methods, as well as hybrid models that combine domain knowledge of mechanistic models with the flexibility of statistical approaches. We also discuss experiences and challenges that arise in the real-world deployment of these forecasting systems, including decision-making informed by forecasts. Finally, we highlight some challenges and open problems found across the forecasting pipeline to enable robust future pandemic preparedness. Forecasting epidemic progression is a complex task influenced by various factors, including human behaviour, pathogen dynamics and environmental conditions. Rodríguez, Kamarthi and colleagues provide a review of machine learning methods for epidemic forecasting from a data-centric computational perspective.","PeriodicalId":48533,"journal":{"name":"Nature Machine Intelligence","volume":"6 10","pages":"1122-1131"},"PeriodicalIF":18.8,"publicationDate":"2024-09-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142325412","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}
Pub Date : 2024-09-26DOI: 10.1038/s42256-024-00903-w
Anthony Bilodeau, Albert Michaud-Gagnon, Julia Chabbert, Benoit Turcotte, Jörn Heine, Audrey Durand, Flavie Lavoie-Cardinal
The integration of artificial intelligence into microscopy systems significantly enhances performance, optimizing both image acquisition and analysis phases. Development of artificial intelligence-assisted super-resolution microscopy is often limited by access to large biological datasets, as well as by difficulties to benchmark and compare approaches on heterogeneous samples. We demonstrate the benefits of a realistic stimulated emission depletion microscopy simulation platform, pySTED, for the development and deployment of artificial intelligence strategies for super-resolution microscopy. pySTED integrates theoretically and empirically validated models for photobleaching and point spread function generation in stimulated emission depletion microscopy, as well as simulating realistic point-scanning dynamics and using a deep learning model to replicate the underlying structures of real images. This simulation environment can be used for data augmentation to train deep neural networks, for the development of online optimization strategies and to train reinforcement learning models. Using pySTED as a training environment allows the reinforcement learning models to bridge the gap between simulation and reality, as showcased by its successful deployment on a real microscope system without fine tuning. Stimulated emission depletion microscopy is a super-resolution imaging technique that utilizes point scanning in fluorescence microscopy. pySTED is developed to aid in the development and benchmarking of optical microscopy experiments, testing it in both synthetic and real settings.
{"title":"Development of AI-assisted microscopy frameworks through realistic simulation with pySTED","authors":"Anthony Bilodeau, Albert Michaud-Gagnon, Julia Chabbert, Benoit Turcotte, Jörn Heine, Audrey Durand, Flavie Lavoie-Cardinal","doi":"10.1038/s42256-024-00903-w","DOIUrl":"10.1038/s42256-024-00903-w","url":null,"abstract":"The integration of artificial intelligence into microscopy systems significantly enhances performance, optimizing both image acquisition and analysis phases. Development of artificial intelligence-assisted super-resolution microscopy is often limited by access to large biological datasets, as well as by difficulties to benchmark and compare approaches on heterogeneous samples. We demonstrate the benefits of a realistic stimulated emission depletion microscopy simulation platform, pySTED, for the development and deployment of artificial intelligence strategies for super-resolution microscopy. pySTED integrates theoretically and empirically validated models for photobleaching and point spread function generation in stimulated emission depletion microscopy, as well as simulating realistic point-scanning dynamics and using a deep learning model to replicate the underlying structures of real images. This simulation environment can be used for data augmentation to train deep neural networks, for the development of online optimization strategies and to train reinforcement learning models. Using pySTED as a training environment allows the reinforcement learning models to bridge the gap between simulation and reality, as showcased by its successful deployment on a real microscope system without fine tuning. Stimulated emission depletion microscopy is a super-resolution imaging technique that utilizes point scanning in fluorescence microscopy. pySTED is developed to aid in the development and benchmarking of optical microscopy experiments, testing it in both synthetic and real settings.","PeriodicalId":48533,"journal":{"name":"Nature Machine Intelligence","volume":"6 10","pages":"1197-1215"},"PeriodicalIF":18.8,"publicationDate":"2024-09-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.nature.com/articles/s42256-024-00903-w.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142321195","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}
Pub Date : 2024-09-25DOI: 10.1038/s42256-024-00897-5
Nick McGreivy, Ammar Hakim
One of the most promising applications of machine learning in computational physics is to accelerate the solution of partial differential equations (PDEs). The key objective of machine-learning-based PDE solvers is to output a sufficiently accurate solution faster than standard numerical methods, which are used as a baseline comparison. We first perform a systematic review of the ML-for-PDE-solving literature. Out of all of the articles that report using ML to solve a fluid-related PDE and claim to outperform a standard numerical method, we determine that 79% (60/76) make a comparison with a weak baseline. Second, we find evidence that reporting biases are widespread, especially outcome reporting and publication biases. We conclude that ML-for-PDE-solving research is overoptimistic: weak baselines lead to overly positive results, while reporting biases lead to under-reporting of negative results. To a large extent, these issues seem to be caused by factors similar to those of past reproducibility crises: researcher degrees of freedom and a bias towards positive results. We call for bottom-up cultural changes to minimize biased reporting as well as top-down structural reforms to reduce perverse incentives for doing so. A systematic review of machine learning approaches to solve partial differential equations related to fluid dynamics highlights concerns about reproducibility and indicates that studies in this area have reached overly optimistic conclusions.
机器学习在计算物理领域最有前途的应用之一是加速偏微分方程(PDE)的求解。基于机器学习的偏微分方程求解器的主要目标是比作为比较基准的标准数值方法更快地输出足够精确的解。我们首先对 ML-for-PDE 求解文献进行了系统回顾。在所有报道使用 ML 解决流体相关 PDE 并声称优于标准数值方法的文章中,我们发现 79% 的文章(60/76)与弱基线进行了比较。其次,我们发现有证据表明报告偏差是普遍存在的,尤其是结果报告和发表偏差。我们的结论是,ML-PDE 仿真研究过于乐观:弱基线导致结果过于乐观,而报告偏差导致负面结果报告不足。在很大程度上,这些问题似乎是由与过去的可重复性危机类似的因素造成的:研究人员的自由度和对积极结果的偏爱。我们呼吁进行自下而上的文化变革,以尽量减少有偏见的报告,同时进行自上而下的结构改革,以减少这样做的不正当激励。
{"title":"Weak baselines and reporting biases lead to overoptimism in machine learning for fluid-related partial differential equations","authors":"Nick McGreivy, Ammar Hakim","doi":"10.1038/s42256-024-00897-5","DOIUrl":"10.1038/s42256-024-00897-5","url":null,"abstract":"One of the most promising applications of machine learning in computational physics is to accelerate the solution of partial differential equations (PDEs). The key objective of machine-learning-based PDE solvers is to output a sufficiently accurate solution faster than standard numerical methods, which are used as a baseline comparison. We first perform a systematic review of the ML-for-PDE-solving literature. Out of all of the articles that report using ML to solve a fluid-related PDE and claim to outperform a standard numerical method, we determine that 79% (60/76) make a comparison with a weak baseline. Second, we find evidence that reporting biases are widespread, especially outcome reporting and publication biases. We conclude that ML-for-PDE-solving research is overoptimistic: weak baselines lead to overly positive results, while reporting biases lead to under-reporting of negative results. To a large extent, these issues seem to be caused by factors similar to those of past reproducibility crises: researcher degrees of freedom and a bias towards positive results. We call for bottom-up cultural changes to minimize biased reporting as well as top-down structural reforms to reduce perverse incentives for doing so. A systematic review of machine learning approaches to solve partial differential equations related to fluid dynamics highlights concerns about reproducibility and indicates that studies in this area have reached overly optimistic conclusions.","PeriodicalId":48533,"journal":{"name":"Nature Machine Intelligence","volume":"6 10","pages":"1256-1269"},"PeriodicalIF":18.8,"publicationDate":"2024-09-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142317761","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}
Biomedical knowledge graphs (KGs) constructed from medical literature have been widely used to validate biomedical discoveries and generate new hypotheses. Recently, large language models (LLMs) have demonstrated a strong ability to generate human-like text data. Although most of these text data have been useful, LLM might also be used to generate malicious content. Here, we investigate whether it is possible that a malicious actor can use an LLM to generate a malicious paper that poisons medical KGs and further affects downstream biomedical applications. As a proof of concept, we develop Scorpius, a conditional text-generation model that generates a malicious paper abstract conditioned on a promoted drug and a target disease. The goal is to fool the medical KG constructed from a mixture of this malicious abstract and millions of real papers so that KG consumers will misidentify this promoted drug as relevant to the target disease. We evaluated Scorpius on a KG constructed from 3,818,528 papers and found that Scorpius can increase the relevance of 71.3% drug–disease pairs from the top 1,000 to the top ten by adding only one malicious abstract. Moreover, the generation of Scorpius achieves better perplexity than ChatGPT, suggesting that such malicious abstracts cannot be efficiently detected by humans. Collectively, Scorpius demonstrates the possibility of poisoning medical KGs and manipulating downstream applications using LLMs, indicating the importance of accountable and trustworthy medical knowledge discovery in the era of LLMs. With increasing reliance on public data sources, researchers are concerned whether low-quality or even adversarial data could have detrimental effects on medical models. Yang et al. developed Scorpius, a malicious text generator, to investigate whether large language models can mislead medical knowledge graphs. They show that a single generated paper abstract can mislead a medical reasoning system that has read millions of papers.
根据医学文献构建的生物医学知识图谱(KGs)已被广泛用于验证生物医学发现和生成新的假设。最近,大型语言模型(LLM)已经证明了生成类人文本数据的强大能力。虽然这些文本数据大多是有用的,但 LLM 也可能被用来生成恶意内容。在此,我们研究了恶意行为者是否有可能利用 LLM 生成恶意论文,从而毒害医学 KG 并进一步影响下游生物医学应用。作为概念验证,我们开发了一个条件文本生成模型 Scorpius,它能根据推广药物和目标疾病生成恶意论文摘要。我们的目标是欺骗由该恶意摘要和数百万真实论文混合构建的医学 KG,使 KG 消费者误认为该促销药物与目标疾病相关。我们在由 3,818,528 篇论文构建的 KG 上对 Scorpius 进行了评估,发现 Scorpius 只需添加一篇恶意摘要,就能将 71.3% 的药物-疾病对的相关性从前 1000 名提高到前 10 名。此外,Scorpius 的生成还比 ChatGPT 获得了更高的困惑度,这表明人类无法有效地检测出此类恶意摘要。总之,Scorpius 证明了利用 LLM 毒化医学 KG 和操纵下游应用程序的可能性,表明了在 LLM 时代负责任和可信赖的医学知识发现的重要性。
{"title":"Poisoning medical knowledge using large language models","authors":"Junwei Yang, Hanwen Xu, Srbuhi Mirzoyan, Tong Chen, Zixuan Liu, Zequn Liu, Wei Ju, Luchen Liu, Zhiping Xiao, Ming Zhang, Sheng Wang","doi":"10.1038/s42256-024-00899-3","DOIUrl":"10.1038/s42256-024-00899-3","url":null,"abstract":"Biomedical knowledge graphs (KGs) constructed from medical literature have been widely used to validate biomedical discoveries and generate new hypotheses. Recently, large language models (LLMs) have demonstrated a strong ability to generate human-like text data. Although most of these text data have been useful, LLM might also be used to generate malicious content. Here, we investigate whether it is possible that a malicious actor can use an LLM to generate a malicious paper that poisons medical KGs and further affects downstream biomedical applications. As a proof of concept, we develop Scorpius, a conditional text-generation model that generates a malicious paper abstract conditioned on a promoted drug and a target disease. The goal is to fool the medical KG constructed from a mixture of this malicious abstract and millions of real papers so that KG consumers will misidentify this promoted drug as relevant to the target disease. We evaluated Scorpius on a KG constructed from 3,818,528 papers and found that Scorpius can increase the relevance of 71.3% drug–disease pairs from the top 1,000 to the top ten by adding only one malicious abstract. Moreover, the generation of Scorpius achieves better perplexity than ChatGPT, suggesting that such malicious abstracts cannot be efficiently detected by humans. Collectively, Scorpius demonstrates the possibility of poisoning medical KGs and manipulating downstream applications using LLMs, indicating the importance of accountable and trustworthy medical knowledge discovery in the era of LLMs. With increasing reliance on public data sources, researchers are concerned whether low-quality or even adversarial data could have detrimental effects on medical models. Yang et al. developed Scorpius, a malicious text generator, to investigate whether large language models can mislead medical knowledge graphs. They show that a single generated paper abstract can mislead a medical reasoning system that has read millions of papers.","PeriodicalId":48533,"journal":{"name":"Nature Machine Intelligence","volume":"6 10","pages":"1156-1168"},"PeriodicalIF":18.8,"publicationDate":"2024-09-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142275878","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}
Pub Date : 2024-09-20DOI: 10.1038/s42256-024-00893-9
Taewi Kim, Insic Hong, Sunghoon Im, Seungeun Rho, Minho Kim, Yeonwook Roh, Changhwan Kim, Jieun Park, Daseul Lim, Doohoe Lee, Seunggon Lee, Jingoo Lee, Inryeol Back, Junggwang Cho, Myung Rae Hong, Sanghun Kang, Joonho Lee, Sungchul Seo, Uikyum Kim, Young-Man Choi, Je-sung Koh, Seungyong Han, Daeshik Kang
Although drone technology has advanced rapidly, replicating the dynamic control and wind-sensing abilities of biological flight is still beyond reach. Biological studies reveal that insect wings are equipped with mechanoreceptors known as campaniform sensilla, which detect complex aerodynamic loads critical for flight agility. By leveraging robotic experiments designed to mimic these biological systems, we confirm that wing strain provides crucial information about the drone’s attitude angle, as well as the direction and velocity of the wind. We introduce a wing-strain-based flight controller that employs the aerodynamic forces exerted on a flapping drone’s wings to deduce vital flight data such as attitude and airflow without accelerometers and gyroscopic sensors. The present work spans five key experiments: initial validation of the wing strain sensor system for state information provision, control in a single degree of freedom movement environment with changing winds, control in a two degrees of freedom movement environment for gravitational attitude adjustment, a test for position control in windy conditions and a demonstration of precise flight path manipulation in a windless condition using only wing strain sensors. We have successfully demonstrated control of a flapping drone in various environments using only wing strain sensors, with the aid of a reinforcement-learning-driven flight controller. The demonstrated adaptability to environmental shifts will be beneficial across varied applications, from gust resistance to wind-assisted flight for autonomous flying robots. Inspired by mechanoreceptors on flying insects, a flapping-wing drone that makes use of strain sensors on the wings and reinforcement-learning-based flight control has been developed. The drone can fly in various unsteady environments, including in windy conditions.
{"title":"Wing-strain-based flight control of flapping-wing drones through reinforcement learning","authors":"Taewi Kim, Insic Hong, Sunghoon Im, Seungeun Rho, Minho Kim, Yeonwook Roh, Changhwan Kim, Jieun Park, Daseul Lim, Doohoe Lee, Seunggon Lee, Jingoo Lee, Inryeol Back, Junggwang Cho, Myung Rae Hong, Sanghun Kang, Joonho Lee, Sungchul Seo, Uikyum Kim, Young-Man Choi, Je-sung Koh, Seungyong Han, Daeshik Kang","doi":"10.1038/s42256-024-00893-9","DOIUrl":"10.1038/s42256-024-00893-9","url":null,"abstract":"Although drone technology has advanced rapidly, replicating the dynamic control and wind-sensing abilities of biological flight is still beyond reach. Biological studies reveal that insect wings are equipped with mechanoreceptors known as campaniform sensilla, which detect complex aerodynamic loads critical for flight agility. By leveraging robotic experiments designed to mimic these biological systems, we confirm that wing strain provides crucial information about the drone’s attitude angle, as well as the direction and velocity of the wind. We introduce a wing-strain-based flight controller that employs the aerodynamic forces exerted on a flapping drone’s wings to deduce vital flight data such as attitude and airflow without accelerometers and gyroscopic sensors. The present work spans five key experiments: initial validation of the wing strain sensor system for state information provision, control in a single degree of freedom movement environment with changing winds, control in a two degrees of freedom movement environment for gravitational attitude adjustment, a test for position control in windy conditions and a demonstration of precise flight path manipulation in a windless condition using only wing strain sensors. We have successfully demonstrated control of a flapping drone in various environments using only wing strain sensors, with the aid of a reinforcement-learning-driven flight controller. The demonstrated adaptability to environmental shifts will be beneficial across varied applications, from gust resistance to wind-assisted flight for autonomous flying robots. Inspired by mechanoreceptors on flying insects, a flapping-wing drone that makes use of strain sensors on the wings and reinforcement-learning-based flight control has been developed. The drone can fly in various unsteady environments, including in windy conditions.","PeriodicalId":48533,"journal":{"name":"Nature Machine Intelligence","volume":"6 9","pages":"992-1005"},"PeriodicalIF":18.8,"publicationDate":"2024-09-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.nature.com/articles/s42256-024-00893-9.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142273337","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}
Pub Date : 2024-09-20DOI: 10.1038/s42256-024-00887-7
Shawn Reeves, Subha Kalyaanamoorthy
Protein sequence likelihood models (PSLMs) are an emerging class of self-supervised deep learning algorithms that learn probability distributions over amino acid identities conditioned on structural or evolutionary context. Recently, PSLMs have demonstrated impressive performance in predicting the relative fitness of variant sequences without any task-specific training, but their potential to address a central goal of protein engineering—enhancing stability—remains underexplored. Here we comprehensively analyse the capacity for zero-shot transfer of eight PSLMs towards prediction of relative thermostability for variants of hundreds of heterogeneous proteins across several quantitative datasets. PSLMs are compared with popular task-specific stability models, and we show that some PSLMs have competitive performance when the appropriate statistics are considered. We highlight relative strengths and weaknesses of PSLMs and examine their complementarity with task-specific models, specifically focusing our analyses on stability-engineering applications. Our results indicate that all PSLMs can appreciably augment the predictions of existing methods by integrating insights from their disparate training objectives, suggesting a path forward in the stagnating field of computational stability prediction. Stabilization of proteins is a key task in protein engineering; however, current methods to predict mutant stability face a number of limitations. Reeves and Kalyaanamoorthy study the performance of self-supervised protein sequence likelihood models for stability prediction and find that combining them with task-specific supervised models can lead to appreciable practical gains.
{"title":"Zero-shot transfer of protein sequence likelihood models to thermostability prediction","authors":"Shawn Reeves, Subha Kalyaanamoorthy","doi":"10.1038/s42256-024-00887-7","DOIUrl":"10.1038/s42256-024-00887-7","url":null,"abstract":"Protein sequence likelihood models (PSLMs) are an emerging class of self-supervised deep learning algorithms that learn probability distributions over amino acid identities conditioned on structural or evolutionary context. Recently, PSLMs have demonstrated impressive performance in predicting the relative fitness of variant sequences without any task-specific training, but their potential to address a central goal of protein engineering—enhancing stability—remains underexplored. Here we comprehensively analyse the capacity for zero-shot transfer of eight PSLMs towards prediction of relative thermostability for variants of hundreds of heterogeneous proteins across several quantitative datasets. PSLMs are compared with popular task-specific stability models, and we show that some PSLMs have competitive performance when the appropriate statistics are considered. We highlight relative strengths and weaknesses of PSLMs and examine their complementarity with task-specific models, specifically focusing our analyses on stability-engineering applications. Our results indicate that all PSLMs can appreciably augment the predictions of existing methods by integrating insights from their disparate training objectives, suggesting a path forward in the stagnating field of computational stability prediction. Stabilization of proteins is a key task in protein engineering; however, current methods to predict mutant stability face a number of limitations. Reeves and Kalyaanamoorthy study the performance of self-supervised protein sequence likelihood models for stability prediction and find that combining them with task-specific supervised models can lead to appreciable practical gains.","PeriodicalId":48533,"journal":{"name":"Nature Machine Intelligence","volume":"6 9","pages":"1063-1076"},"PeriodicalIF":18.8,"publicationDate":"2024-09-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142273333","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}