Pub Date : 2024-08-29DOI: 10.1038/s42256-024-00886-8
Emanuele Zappala, Antonio Henrique de Oliveira Fonseca, Josue Ortega Caro, Andrew Henry Moberly, Michael James Higley, Jessica Cardin, David van Dijk
Nonlinear operators with long-distance spatiotemporal dependencies are fundamental in modelling complex systems across sciences; yet, learning these non-local operators remains challenging in machine learning. Integral equations, which model such non-local systems, have wide-ranging applications in physics, chemistry, biology and engineering. We introduce the neural integral equation, a method for learning unknown integral operators from data using an integral equation solver. To improve scalability and model capacity, we also present the attentional neural integral equation, which replaces the integral with self-attention. Both models are grounded in the theory of second-kind integral equations, where the indeterminate appears both inside and outside the integral operator. We provide a theoretical analysis showing how self-attention can approximate integral operators under mild regularity assumptions, further deepening previously reported connections between transformers and integration, as well as deriving corresponding approximation results for integral operators. Through numerical benchmarks on synthetic and real-world data, including Lotka–Volterra, Navier–Stokes and Burgers’ equations, as well as brain dynamics and integral equations, we showcase the models’ capabilities and their ability to derive interpretable dynamics embeddings. Our experiments demonstrate that attentional neural integral equations outperform existing methods, especially for longer time intervals and higher-dimensional problems. Our work addresses a critical gap in machine learning for non-local operators and offers a powerful tool for studying unknown complex systems with long-range dependencies. Integral equations are used in science and engineering to model complex systems with non-local dependencies; however, existing traditional and machine-learning-based methods cannot yield accurate or efficient solutions in several complex cases. Zappala and colleagues introduce a neural-network-based method that can learn an integral operator and its dynamics from data, demonstrating higher accuracy or scalability compared with several state-of-the-art methods.
{"title":"Learning integral operators via neural integral equations","authors":"Emanuele Zappala, Antonio Henrique de Oliveira Fonseca, Josue Ortega Caro, Andrew Henry Moberly, Michael James Higley, Jessica Cardin, David van Dijk","doi":"10.1038/s42256-024-00886-8","DOIUrl":"10.1038/s42256-024-00886-8","url":null,"abstract":"Nonlinear operators with long-distance spatiotemporal dependencies are fundamental in modelling complex systems across sciences; yet, learning these non-local operators remains challenging in machine learning. Integral equations, which model such non-local systems, have wide-ranging applications in physics, chemistry, biology and engineering. We introduce the neural integral equation, a method for learning unknown integral operators from data using an integral equation solver. To improve scalability and model capacity, we also present the attentional neural integral equation, which replaces the integral with self-attention. Both models are grounded in the theory of second-kind integral equations, where the indeterminate appears both inside and outside the integral operator. We provide a theoretical analysis showing how self-attention can approximate integral operators under mild regularity assumptions, further deepening previously reported connections between transformers and integration, as well as deriving corresponding approximation results for integral operators. Through numerical benchmarks on synthetic and real-world data, including Lotka–Volterra, Navier–Stokes and Burgers’ equations, as well as brain dynamics and integral equations, we showcase the models’ capabilities and their ability to derive interpretable dynamics embeddings. Our experiments demonstrate that attentional neural integral equations outperform existing methods, especially for longer time intervals and higher-dimensional problems. Our work addresses a critical gap in machine learning for non-local operators and offers a powerful tool for studying unknown complex systems with long-range dependencies. Integral equations are used in science and engineering to model complex systems with non-local dependencies; however, existing traditional and machine-learning-based methods cannot yield accurate or efficient solutions in several complex cases. Zappala and colleagues introduce a neural-network-based method that can learn an integral operator and its dynamics from data, demonstrating higher accuracy or scalability compared with several state-of-the-art methods.","PeriodicalId":48533,"journal":{"name":"Nature Machine Intelligence","volume":"6 9","pages":"1046-1062"},"PeriodicalIF":18.8,"publicationDate":"2024-08-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.nature.com/articles/s42256-024-00886-8.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142090186","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 : 2024-08-27DOI: 10.1038/s42256-024-00888-6
Yi Zhong, Gaozheng Li, Ji Yang, Houbing Zheng, Yongqiang Yu, Jiheng Zhang, Heng Luo, Biao Wang, Zuquan Weng
Unexpected drug–drug interactions (DDIs) are important issues for both pharmaceutical research and clinical applications due to the high risk of causing severe adverse drug reactions or drug withdrawals. Many deep learning models have achieved high performance in DDI prediction, but model interpretability to reveal the underlying causes of DDIs has not been extensively explored. Here we propose MeTDDI—a deep learning framework with local–global self-attention and co-attention to learn motif-based graphs for DDI prediction. MeTDDI achieved competitive performance compared with state-of-the-art models. Regarding interpretability, we conducted extensive assessments on 73 drugs with 13,786 DDIs and MeTDDI can precisely explain the structural mechanisms for 5,602 DDIs involving 58 drugs. Besides, MeTDDI shows potential to explain complex DDI mechanisms and mitigate DDI risks. To summarize, MeTDDI provides a new perspective on exploring DDI mechanisms, which will benefit both drug discovery and polypharmacy for safer therapies for patients. A transformer-based approach that predicts drug–drug interactions in polypharmacy has been shown, which also identifies perpetrator drugs and the chemical mechanisms causing the interactions. The method could facilitate high-throughput optimization of drug combinations and mitigate adverse drug–drug interaction risks.
{"title":"Learning motif-based graphs for drug–drug interaction prediction via local–global self-attention","authors":"Yi Zhong, Gaozheng Li, Ji Yang, Houbing Zheng, Yongqiang Yu, Jiheng Zhang, Heng Luo, Biao Wang, Zuquan Weng","doi":"10.1038/s42256-024-00888-6","DOIUrl":"10.1038/s42256-024-00888-6","url":null,"abstract":"Unexpected drug–drug interactions (DDIs) are important issues for both pharmaceutical research and clinical applications due to the high risk of causing severe adverse drug reactions or drug withdrawals. Many deep learning models have achieved high performance in DDI prediction, but model interpretability to reveal the underlying causes of DDIs has not been extensively explored. Here we propose MeTDDI—a deep learning framework with local–global self-attention and co-attention to learn motif-based graphs for DDI prediction. MeTDDI achieved competitive performance compared with state-of-the-art models. Regarding interpretability, we conducted extensive assessments on 73 drugs with 13,786 DDIs and MeTDDI can precisely explain the structural mechanisms for 5,602 DDIs involving 58 drugs. Besides, MeTDDI shows potential to explain complex DDI mechanisms and mitigate DDI risks. To summarize, MeTDDI provides a new perspective on exploring DDI mechanisms, which will benefit both drug discovery and polypharmacy for safer therapies for patients. A transformer-based approach that predicts drug–drug interactions in polypharmacy has been shown, which also identifies perpetrator drugs and the chemical mechanisms causing the interactions. The method could facilitate high-throughput optimization of drug combinations and mitigate adverse drug–drug interaction risks.","PeriodicalId":48533,"journal":{"name":"Nature Machine Intelligence","volume":"6 9","pages":"1094-1105"},"PeriodicalIF":18.8,"publicationDate":"2024-08-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142085189","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 : 2024-08-27DOI: 10.1038/s42256-024-00883-x
Davide Carnevali, Limei Zhong, Esther González-Almela, Carlotta Viana, Mikhail Rotkevich, Aiping Wang, Daniel Franco-Barranco, Aitor Gonzalez-Marfil, Maria Victoria Neguembor, Alvaro Castells-Garcia, Ignacio Arganda-Carreras, Maria Pia Cosma
Cellular phenotypic heterogeneity is an important hallmark of many biological processes and understanding its origins remains a substantial challenge. This heterogeneity often reflects variations in the chromatin structure, influenced by factors such as viral infections and cancer, which dramatically reshape the cellular landscape. To address the challenge of identifying distinct cell states, we developed artificial intelligence of the nucleus (AINU), a deep learning method that can identify specific nuclear signatures at the nanoscale resolution. AINU can distinguish different cell states based on the spatial arrangement of core histone H3, RNA polymerase II or DNA from super-resolution microscopy images. With only a small number of images as the training data, AINU correctly identifies human somatic cells, human-induced pluripotent stem cells, very early stage infected cells transduced with DNA herpes simplex virus type 1 and even cancer cells after appropriate retraining. Finally, using AI interpretability methods, we find that the RNA polymerase II localizations in the nucleoli aid in distinguishing human-induced pluripotent stem cells from their somatic cells. Overall, AINU coupled with super-resolution microscopy of nuclear structures provides a robust tool for the precise detection of cellular heterogeneity, with considerable potential for advancing diagnostics and therapies in regenerative medicine, virology and cancer biology. Cellular phenotypic heterogeneity is a key determinant of biological functions and is challenging to identify. A deep learning method that recognizes specific nuclear signatures is discussed, which can identify cellular heterogeneity and differentiate between various cell states using a small amount of super-resolution microscopy data.
细胞表型异质性是许多生物过程的重要标志，而了解其起源仍是一项巨大的挑战。这种异质性通常反映了染色质结构的变化，受病毒感染和癌症等因素的影响，这些因素极大地重塑了细胞景观。为了应对识别不同细胞状态的挑战，我们开发了细胞核人工智能（AINU），这是一种深度学习方法，可以在纳米级分辨率上识别特定的核特征。AINU 可以根据超分辨率显微镜图像中核心组蛋白 H3、RNA 聚合酶 II 或 DNA 的空间排列来区分不同的细胞状态。AINU 仅用少量图像作为训练数据，就能正确识别人类体细胞、人类诱导多能干细胞、DNA 1 型单纯疱疹病毒转导的早期感染细胞，甚至在经过适当的再训练后还能识别癌细胞。最后，利用人工智能可解释性方法，我们发现核小体中 RNA 聚合酶 II 的定位有助于区分人类诱导多能干细胞和体细胞。总之，AINU 与核结构超分辨率显微镜相结合，为精确检测细胞异质性提供了一个强大的工具，在再生医学、病毒学和癌症生物学的诊断和治疗方面具有相当大的潜力。
{"title":"A deep learning method that identifies cellular heterogeneity using nanoscale nuclear features","authors":"Davide Carnevali, Limei Zhong, Esther González-Almela, Carlotta Viana, Mikhail Rotkevich, Aiping Wang, Daniel Franco-Barranco, Aitor Gonzalez-Marfil, Maria Victoria Neguembor, Alvaro Castells-Garcia, Ignacio Arganda-Carreras, Maria Pia Cosma","doi":"10.1038/s42256-024-00883-x","DOIUrl":"10.1038/s42256-024-00883-x","url":null,"abstract":"Cellular phenotypic heterogeneity is an important hallmark of many biological processes and understanding its origins remains a substantial challenge. This heterogeneity often reflects variations in the chromatin structure, influenced by factors such as viral infections and cancer, which dramatically reshape the cellular landscape. To address the challenge of identifying distinct cell states, we developed artificial intelligence of the nucleus (AINU), a deep learning method that can identify specific nuclear signatures at the nanoscale resolution. AINU can distinguish different cell states based on the spatial arrangement of core histone H3, RNA polymerase II or DNA from super-resolution microscopy images. With only a small number of images as the training data, AINU correctly identifies human somatic cells, human-induced pluripotent stem cells, very early stage infected cells transduced with DNA herpes simplex virus type 1 and even cancer cells after appropriate retraining. Finally, using AI interpretability methods, we find that the RNA polymerase II localizations in the nucleoli aid in distinguishing human-induced pluripotent stem cells from their somatic cells. Overall, AINU coupled with super-resolution microscopy of nuclear structures provides a robust tool for the precise detection of cellular heterogeneity, with considerable potential for advancing diagnostics and therapies in regenerative medicine, virology and cancer biology. Cellular phenotypic heterogeneity is a key determinant of biological functions and is challenging to identify. A deep learning method that recognizes specific nuclear signatures is discussed, which can identify cellular heterogeneity and differentiate between various cell states using a small amount of super-resolution microscopy data.","PeriodicalId":48533,"journal":{"name":"Nature Machine Intelligence","volume":"6 9","pages":"1021-1033"},"PeriodicalIF":18.8,"publicationDate":"2024-08-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.nature.com/articles/s42256-024-00883-x.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142085188","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 : 2024-08-22DOI: 10.1038/s42256-024-00881-z
Isabelle Augenstein, Timothy Baldwin, Meeyoung Cha, Tanmoy Chakraborty, Giovanni Luca Ciampaglia, David Corney, Renee DiResta, Emilio Ferrara, Scott Hale, Alon Halevy, Eduard Hovy, Heng Ji, Filippo Menczer, Ruben Miguez, Preslav Nakov, Dietram Scheufele, Shivam Sharma, Giovanni Zagni
The emergence of tools based on large language models (LLMs), such as OpenAI’s ChatGPT and Google’s Gemini, has garnered immense public attention owing to their advanced natural language generation capabilities. These remarkably natural-sounding tools have the potential to be highly useful for various tasks. However, they also tend to produce false, erroneous or misleading content—commonly referred to as hallucinations. Moreover, LLMs can be misused to generate convincing, yet false, content and profiles on a large scale, posing a substantial societal challenge by potentially deceiving users and spreading inaccurate information. This makes fact-checking increasingly important. Despite their issues with factual accuracy, LLMs have shown proficiency in various subtasks that support fact-checking, which is essential to ensure factually accurate responses. In light of these concerns, we explore issues related to factuality in LLMs and their impact on fact-checking. We identify key challenges, imminent threats and possible solutions to these factuality issues. We also thoroughly examine these challenges, existing solutions and potential prospects for fact-checking. By analysing the factuality constraints within LLMs and their impact on fact-checking, we aim to contribute to a path towards maintaining accuracy at a time of confluence of generative artificial intelligence and misinformation. Large language models (LLMs) present challenges, including a tendency to produce false or misleading content and the potential to create misinformation or disinformation. Augenstein and colleagues explore issues related to factuality in LLMs and their impact on fact-checking.
{"title":"Factuality challenges in the era of large language models and opportunities for fact-checking","authors":"Isabelle Augenstein, Timothy Baldwin, Meeyoung Cha, Tanmoy Chakraborty, Giovanni Luca Ciampaglia, David Corney, Renee DiResta, Emilio Ferrara, Scott Hale, Alon Halevy, Eduard Hovy, Heng Ji, Filippo Menczer, Ruben Miguez, Preslav Nakov, Dietram Scheufele, Shivam Sharma, Giovanni Zagni","doi":"10.1038/s42256-024-00881-z","DOIUrl":"10.1038/s42256-024-00881-z","url":null,"abstract":"The emergence of tools based on large language models (LLMs), such as OpenAI’s ChatGPT and Google’s Gemini, has garnered immense public attention owing to their advanced natural language generation capabilities. These remarkably natural-sounding tools have the potential to be highly useful for various tasks. However, they also tend to produce false, erroneous or misleading content—commonly referred to as hallucinations. Moreover, LLMs can be misused to generate convincing, yet false, content and profiles on a large scale, posing a substantial societal challenge by potentially deceiving users and spreading inaccurate information. This makes fact-checking increasingly important. Despite their issues with factual accuracy, LLMs have shown proficiency in various subtasks that support fact-checking, which is essential to ensure factually accurate responses. In light of these concerns, we explore issues related to factuality in LLMs and their impact on fact-checking. We identify key challenges, imminent threats and possible solutions to these factuality issues. We also thoroughly examine these challenges, existing solutions and potential prospects for fact-checking. By analysing the factuality constraints within LLMs and their impact on fact-checking, we aim to contribute to a path towards maintaining accuracy at a time of confluence of generative artificial intelligence and misinformation. Large language models (LLMs) present challenges, including a tendency to produce false or misleading content and the potential to create misinformation or disinformation. Augenstein and colleagues explore issues related to factuality in LLMs and their impact on fact-checking.","PeriodicalId":48533,"journal":{"name":"Nature Machine Intelligence","volume":"6 8","pages":"852-863"},"PeriodicalIF":18.8,"publicationDate":"2024-08-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142022075","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 : 2024-08-14DOI: 10.1038/s42256-024-00876-w
Bin Feng, Zequn Liu, Nanlan Huang, Zhiping Xiao, Haomiao Zhang, Srbuhi Mirzoyan, Hanwen Xu, Jiaran Hao, Yinghui Xu, Ming Zhang, Sheng Wang
The bioactivity of compounds plays an important role in drug development and discovery. Existing machine learning approaches have poor generalizability in bioactivity prediction due to the small number of compounds in each assay and incompatible measurements among assays. In this paper, we propose ActFound, a bioactivity foundation model trained on 1.6 million experimentally measured bioactivities and 35,644 assays from ChEMBL. The key idea of ActFound is to use pairwise learning to learn the relative bioactivity differences between two compounds within the same assay to circumvent the incompatibility among assays. ActFound further exploits meta-learning to jointly optimize the model from all assays. On six real-world bioactivity datasets, ActFound demonstrates accurate in-domain prediction and strong generalization across assay types and molecular scaffolds. We also demonstrate that ActFound can be used as an accurate alternative to the leading physics-based computational tool FEP+(OPLS4) by achieving comparable performance when using only a few data points for fine-tuning. Our promising results indicate that ActFound could be an effective bioactivity foundation model for compound bioactivity prediction, paving the way for machine-learning-based drug development and discovery. Traditional machine learning methods for drug development struggle with bioactivity prediction due to the limited number of compounds in each assay and assay incompatibilities. Feng et al. developed ActFound, a bioactivity foundation model trained by pairwise learning and meta-learning, that improves the accuracy and generalization of bioactivity prediction.
{"title":"A bioactivity foundation model using pairwise meta-learning","authors":"Bin Feng, Zequn Liu, Nanlan Huang, Zhiping Xiao, Haomiao Zhang, Srbuhi Mirzoyan, Hanwen Xu, Jiaran Hao, Yinghui Xu, Ming Zhang, Sheng Wang","doi":"10.1038/s42256-024-00876-w","DOIUrl":"10.1038/s42256-024-00876-w","url":null,"abstract":"The bioactivity of compounds plays an important role in drug development and discovery. Existing machine learning approaches have poor generalizability in bioactivity prediction due to the small number of compounds in each assay and incompatible measurements among assays. In this paper, we propose ActFound, a bioactivity foundation model trained on 1.6 million experimentally measured bioactivities and 35,644 assays from ChEMBL. The key idea of ActFound is to use pairwise learning to learn the relative bioactivity differences between two compounds within the same assay to circumvent the incompatibility among assays. ActFound further exploits meta-learning to jointly optimize the model from all assays. On six real-world bioactivity datasets, ActFound demonstrates accurate in-domain prediction and strong generalization across assay types and molecular scaffolds. We also demonstrate that ActFound can be used as an accurate alternative to the leading physics-based computational tool FEP+(OPLS4) by achieving comparable performance when using only a few data points for fine-tuning. Our promising results indicate that ActFound could be an effective bioactivity foundation model for compound bioactivity prediction, paving the way for machine-learning-based drug development and discovery. Traditional machine learning methods for drug development struggle with bioactivity prediction due to the limited number of compounds in each assay and assay incompatibilities. Feng et al. developed ActFound, a bioactivity foundation model trained by pairwise learning and meta-learning, that improves the accuracy and generalization of bioactivity prediction.","PeriodicalId":48533,"journal":{"name":"Nature Machine Intelligence","volume":"6 8","pages":"962-974"},"PeriodicalIF":18.8,"publicationDate":"2024-08-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141980824","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 : 2024-08-12DOI: 10.1038/s42256-024-00874-y
Surbhi Mittal, Kartik Thakral, Richa Singh, Mayank Vatsa, Tamar Glaser, Cristian Canton Ferrer, Tal Hassner
Artificial Intelligence (AI) has seamlessly integrated into numerous scientific domains, catalysing unparalleled enhancements across a broad spectrum of tasks; however, its integrity and trustworthiness have emerged as notable concerns. The scientific community has focused on the development of trustworthy AI algorithms; however, machine learning and deep learning algorithms, popular in the AI community today, intrinsically rely on the quality of their training data. These algorithms are designed to detect patterns within the data, thereby learning the intended behavioural objectives. Any inadequacy in the data has the potential to translate directly into algorithms. In this study we discuss the importance of responsible machine learning datasets through the lens of fairness, privacy and regulatory compliance, and present a large audit of computer vision datasets. Despite the ubiquity of fairness and privacy challenges across diverse data domains, current regulatory frameworks primarily address human-centric data concerns. We therefore focus our discussion on biometric and healthcare datasets, although the principles we outline are broadly applicable across various domains. The audit is conducted through evaluation of the proposed responsible rubric. After surveying over 100 datasets, our detailed analysis of 60 distinct datasets highlights a universal susceptibility to fairness, privacy and regulatory compliance issues. This finding emphasizes the urgent need for revising dataset creation methodologies within the scientific community, especially in light of global advancements in data protection legislation. We assert that our study is critically relevant in the contemporary AI context, offering insights and recommendations that are both timely and essential for the ongoing evolution of AI technologies. There are pervasive concerns related to fairness, privacy and regulatory compliance in machine learning applications in healthcare, necessitating a reevaluation of dataset creation practices. Mittal et al. examine various computer vision datasets, providing insights to foster responsible AI development.
{"title":"On responsible machine learning datasets emphasizing fairness, privacy and regulatory norms with examples in biometrics and healthcare","authors":"Surbhi Mittal, Kartik Thakral, Richa Singh, Mayank Vatsa, Tamar Glaser, Cristian Canton Ferrer, Tal Hassner","doi":"10.1038/s42256-024-00874-y","DOIUrl":"10.1038/s42256-024-00874-y","url":null,"abstract":"Artificial Intelligence (AI) has seamlessly integrated into numerous scientific domains, catalysing unparalleled enhancements across a broad spectrum of tasks; however, its integrity and trustworthiness have emerged as notable concerns. The scientific community has focused on the development of trustworthy AI algorithms; however, machine learning and deep learning algorithms, popular in the AI community today, intrinsically rely on the quality of their training data. These algorithms are designed to detect patterns within the data, thereby learning the intended behavioural objectives. Any inadequacy in the data has the potential to translate directly into algorithms. In this study we discuss the importance of responsible machine learning datasets through the lens of fairness, privacy and regulatory compliance, and present a large audit of computer vision datasets. Despite the ubiquity of fairness and privacy challenges across diverse data domains, current regulatory frameworks primarily address human-centric data concerns. We therefore focus our discussion on biometric and healthcare datasets, although the principles we outline are broadly applicable across various domains. The audit is conducted through evaluation of the proposed responsible rubric. After surveying over 100 datasets, our detailed analysis of 60 distinct datasets highlights a universal susceptibility to fairness, privacy and regulatory compliance issues. This finding emphasizes the urgent need for revising dataset creation methodologies within the scientific community, especially in light of global advancements in data protection legislation. We assert that our study is critically relevant in the contemporary AI context, offering insights and recommendations that are both timely and essential for the ongoing evolution of AI technologies. There are pervasive concerns related to fairness, privacy and regulatory compliance in machine learning applications in healthcare, necessitating a reevaluation of dataset creation practices. Mittal et al. examine various computer vision datasets, providing insights to foster responsible AI development.","PeriodicalId":48533,"journal":{"name":"Nature Machine Intelligence","volume":"6 8","pages":"936-949"},"PeriodicalIF":18.8,"publicationDate":"2024-08-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.nature.com/articles/s42256-024-00874-y.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141974035","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 : 2024-08-09DOI: 10.1038/s42256-024-00882-y
Mark Endo, Favour Nerrise, Qingyu Zhao, Edith V. Sullivan, Li Fei-Fei, Victor W. Henderson, Kilian M. Pohl, Kathleen L. Poston, Ehsan Adeli
Neurodegenerative diseases manifest different motor and cognitive signs and symptoms that are highly heterogeneous. Parsing these heterogeneities may lead to an improved understanding of underlying disease mechanisms; however, current methods are dependent on clinical assessments and an arbitrary choice of behavioural tests. Here we present a data-driven subtyping approach using video-captured human motion and brain functional connectivity from resting-state functional magnetic resonance imaging. We applied our framework to a cohort of individuals at different stages of Parkinson’s disease. The process mapped the data to low-dimensional measures by projecting them onto a canonical correlation space that identified three Parkinson’s disease subtypes: subtype I was characterized by motor difficulties and poor visuospatial abilities; subtype II exhibited difficulties in non-motor components of activities of daily living and motor complications (dyskinesias and motor fluctuations) and subtype III was characterized by predominant tremor symptoms. We conducted a convergent validity analysis by comparing our approach to existing and widely used approaches. The compared approaches yielded subtypes that were adequately well-clustered in the motion-brain representation space we created to delineate subtypes. Our data-driven approach, contrary to other forms of subtyping, derived biomarkers predictive of motion impairment and subtype memberships that were captured objectively by digital videos. Diagnostic strategies for neurodegenerative diseases involve various data types, related to motor and cognitive signals. Endo et al. describe a data-driven subtyping approach for Parkinson’s disease, combining motion data (from videos) and brain functional connectivity data. The method reveals clinically relevant subtypes and digital biomarkers, uncovering movement-linked heterogeneities of Parkinson’s disease.
神经退行性疾病表现出不同的运动和认知体征和症状，具有高度异质性。解析这些异质性可能有助于更好地了解潜在的疾病机制；然而，目前的方法依赖于临床评估和任意选择的行为测试。在这里，我们提出了一种数据驱动的亚型分析方法，该方法使用视频捕捉的人体运动和静息状态功能磁共振成像的大脑功能连接。我们将这一框架应用于处于帕金森病不同阶段的人群。这一过程通过将数据投射到一个典型相关空间，将数据映射到低维测量值，从而确定了三种帕金森病亚型：亚型 I 的特征是运动困难和视觉空间能力差；亚型 II 表现出日常生活活动中的非运动部分困难和运动并发症（运动障碍和运动波动）；亚型 III 的特征是震颤症状占主导地位。通过将我们的方法与现有的、广泛使用的方法进行比较，我们进行了收敛有效性分析。比较后得出的亚型在我们为划分亚型而创建的运动-大脑表征空间中充分聚类。与其他形式的亚型划分方法不同，我们的数据驱动方法通过数字视频客观地捕捉到了可预测运动障碍和亚型成员的生物标志物。
{"title":"Data-driven discovery of movement-linked heterogeneity in neurodegenerative diseases","authors":"Mark Endo, Favour Nerrise, Qingyu Zhao, Edith V. Sullivan, Li Fei-Fei, Victor W. Henderson, Kilian M. Pohl, Kathleen L. Poston, Ehsan Adeli","doi":"10.1038/s42256-024-00882-y","DOIUrl":"10.1038/s42256-024-00882-y","url":null,"abstract":"Neurodegenerative diseases manifest different motor and cognitive signs and symptoms that are highly heterogeneous. Parsing these heterogeneities may lead to an improved understanding of underlying disease mechanisms; however, current methods are dependent on clinical assessments and an arbitrary choice of behavioural tests. Here we present a data-driven subtyping approach using video-captured human motion and brain functional connectivity from resting-state functional magnetic resonance imaging. We applied our framework to a cohort of individuals at different stages of Parkinson’s disease. The process mapped the data to low-dimensional measures by projecting them onto a canonical correlation space that identified three Parkinson’s disease subtypes: subtype I was characterized by motor difficulties and poor visuospatial abilities; subtype II exhibited difficulties in non-motor components of activities of daily living and motor complications (dyskinesias and motor fluctuations) and subtype III was characterized by predominant tremor symptoms. We conducted a convergent validity analysis by comparing our approach to existing and widely used approaches. The compared approaches yielded subtypes that were adequately well-clustered in the motion-brain representation space we created to delineate subtypes. Our data-driven approach, contrary to other forms of subtyping, derived biomarkers predictive of motion impairment and subtype memberships that were captured objectively by digital videos. Diagnostic strategies for neurodegenerative diseases involve various data types, related to motor and cognitive signals. Endo et al. describe a data-driven subtyping approach for Parkinson’s disease, combining motion data (from videos) and brain functional connectivity data. The method reveals clinically relevant subtypes and digital biomarkers, uncovering movement-linked heterogeneities of Parkinson’s disease.","PeriodicalId":48533,"journal":{"name":"Nature Machine Intelligence","volume":"6 9","pages":"1034-1045"},"PeriodicalIF":18.8,"publicationDate":"2024-08-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141908969","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 : 2024-08-08DOI: 10.1038/s42256-024-00885-9
Margaret C. von Ebers, Xue-Xin Wei
Constructing spatial maps from sensory inputs is challenging in both neuroscience and artificial intelligence. A recent study demonstrates that a self-attention neural network using predictive coding can generate an environmental map in its latent space as an agent that navigates the environment.
{"title":"Cognitive maps from predictive vision","authors":"Margaret C. von Ebers, Xue-Xin Wei","doi":"10.1038/s42256-024-00885-9","DOIUrl":"10.1038/s42256-024-00885-9","url":null,"abstract":"Constructing spatial maps from sensory inputs is challenging in both neuroscience and artificial intelligence. A recent study demonstrates that a self-attention neural network using predictive coding can generate an environmental map in its latent space as an agent that navigates the environment.","PeriodicalId":48533,"journal":{"name":"Nature Machine Intelligence","volume":"6 8","pages":"850-851"},"PeriodicalIF":18.8,"publicationDate":"2024-08-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141904419","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 : 2024-08-05DOI: 10.1038/s42256-024-00873-z
Shihao Feng, Zhenyu Chen, Chengwei Zhang, Yuhao Xie, Sergey Ovchinnikov, Yi Qin Gao, Sirui Liu
Protein complex structure prediction plays important roles in various applications, such as drug discovery and antibody design. However, due to limited prediction accuracy, there are frequent inconsistencies between the predictions and the experiments. Here we present ColabDock, a general framework adapting deep learning structure prediction models to integrate experimental restraints of different forms and sources without further large-scale retraining or fine tuning. With a generation–prediction architecture and trained ranking model, ColabDock outperforms HADDOCK and ClusPro using AlphaFold2 as the structure prediction model, not only in complex structure predictions with simulated residue and surface restraints but also in those assisted by nuclear magnetic resonance chemical shift perturbation as well as covalent labelling. It also assists antibody–antigen interface prediction with emulated interface scan restraints, which could be obtained by experiments such as deep mutational scanning. As a unified framework, we hope that ColabDock can help to bridge the gap between experimental and computational protein science. Despite rapid developments in predicting the complex structures of proteins, there are still inconsistencies between predictions and experiments. Feng et al. developed ColabDock, a general framework for deep learning models that integrates various experimental restraints and improves complex interface prediction, including antibody–antigen interactions.
{"title":"Integrated structure prediction of protein–protein docking with experimental restraints using ColabDock","authors":"Shihao Feng, Zhenyu Chen, Chengwei Zhang, Yuhao Xie, Sergey Ovchinnikov, Yi Qin Gao, Sirui Liu","doi":"10.1038/s42256-024-00873-z","DOIUrl":"10.1038/s42256-024-00873-z","url":null,"abstract":"Protein complex structure prediction plays important roles in various applications, such as drug discovery and antibody design. However, due to limited prediction accuracy, there are frequent inconsistencies between the predictions and the experiments. Here we present ColabDock, a general framework adapting deep learning structure prediction models to integrate experimental restraints of different forms and sources without further large-scale retraining or fine tuning. With a generation–prediction architecture and trained ranking model, ColabDock outperforms HADDOCK and ClusPro using AlphaFold2 as the structure prediction model, not only in complex structure predictions with simulated residue and surface restraints but also in those assisted by nuclear magnetic resonance chemical shift perturbation as well as covalent labelling. It also assists antibody–antigen interface prediction with emulated interface scan restraints, which could be obtained by experiments such as deep mutational scanning. As a unified framework, we hope that ColabDock can help to bridge the gap between experimental and computational protein science. Despite rapid developments in predicting the complex structures of proteins, there are still inconsistencies between predictions and experiments. Feng et al. developed ColabDock, a general framework for deep learning models that integrates various experimental restraints and improves complex interface prediction, including antibody–antigen interactions.","PeriodicalId":48533,"journal":{"name":"Nature Machine Intelligence","volume":"6 8","pages":"924-935"},"PeriodicalIF":18.8,"publicationDate":"2024-08-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141895210","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 : 2024-08-02DOI: 10.1038/s42256-024-00890-y
Yanyi Chu, Dan Yu, Yupeng Li, Kaixuan Huang, Yue Shen, Le Cong, Jason Zhang, Mengdi Wang
{"title":"Author Correction: A 5′ UTR language model for decoding untranslated regions of mRNA and function predictions","authors":"Yanyi Chu, Dan Yu, Yupeng Li, Kaixuan Huang, Yue Shen, Le Cong, Jason Zhang, Mengdi Wang","doi":"10.1038/s42256-024-00890-y","DOIUrl":"10.1038/s42256-024-00890-y","url":null,"abstract":"","PeriodicalId":48533,"journal":{"name":"Nature Machine Intelligence","volume":"6 8","pages":"988-988"},"PeriodicalIF":18.8,"publicationDate":"2024-08-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.nature.com/articles/s42256-024-00890-y.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142091165","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}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：