Pub Date : 2025-01-13DOI: 10.1038/s42256-024-00969-6
Liwei Jiang, Jena D. Hwang, Chandra Bhagavatula, Ronan Le Bras, Jenny T. Liang, Sydney Levine, Jesse Dodge, Keisuke Sakaguchi, Maxwell Forbes, Jack Hessel, Jon Borchardt, Taylor Sorensen, Saadia Gabriel, Yulia Tsvetkov, Oren Etzioni, Maarten Sap, Regina Rini, Yejin Choi
As our society adopts increasingly powerful artificial intelligence (AI) systems for pervasive use, there are growing concerns about machine morality—or lack thereof. Millions of users already rely on the outputs of AI systems, such as chatbots, as decision aids. Meanwhile, AI researchers continue to grapple with the challenge of aligning these systems with human morality and values. In response to this challenge, we build and test Delphi, an open-source AI system trained to predict the moral judgements of US participants. The computational framework of Delphi is grounded in the framework proposed by the prominent moral philosopher John Rawls. Our results speak to the promises and limits of teaching machines about human morality. Delphi demonstrates improved generalization capabilities over those exhibited by off-the-shelf neural language models. At the same time, Delphi’s failures also underscore important challenges in this arena. For instance, Delphi has limited cultural awareness and is susceptible to pervasive biases. Despite these shortcomings, we demonstrate several compelling use cases of Delphi, including its incorporation as a component within an ensemble of AI systems. Finally, we computationally demonstrate the potential of Rawls’s prospect of hybrid approaches for reliable moral reasoning, inspiring future research in computational morality. Aligning artificial intelligence systems with human morality poses scientific, societal and ethical challenges. Delphi, an artificial intelligence system designed to predict human moral judgements based on John Rawls’s philosophical framework, is developed and tested, highlighting its potential for ethical applications and emphasizing the need to address its limitations and biases.
{"title":"Investigating machine moral judgement through the Delphi experiment","authors":"Liwei Jiang, Jena D. Hwang, Chandra Bhagavatula, Ronan Le Bras, Jenny T. Liang, Sydney Levine, Jesse Dodge, Keisuke Sakaguchi, Maxwell Forbes, Jack Hessel, Jon Borchardt, Taylor Sorensen, Saadia Gabriel, Yulia Tsvetkov, Oren Etzioni, Maarten Sap, Regina Rini, Yejin Choi","doi":"10.1038/s42256-024-00969-6","DOIUrl":"10.1038/s42256-024-00969-6","url":null,"abstract":"As our society adopts increasingly powerful artificial intelligence (AI) systems for pervasive use, there are growing concerns about machine morality—or lack thereof. Millions of users already rely on the outputs of AI systems, such as chatbots, as decision aids. Meanwhile, AI researchers continue to grapple with the challenge of aligning these systems with human morality and values. In response to this challenge, we build and test Delphi, an open-source AI system trained to predict the moral judgements of US participants. The computational framework of Delphi is grounded in the framework proposed by the prominent moral philosopher John Rawls. Our results speak to the promises and limits of teaching machines about human morality. Delphi demonstrates improved generalization capabilities over those exhibited by off-the-shelf neural language models. At the same time, Delphi’s failures also underscore important challenges in this arena. For instance, Delphi has limited cultural awareness and is susceptible to pervasive biases. Despite these shortcomings, we demonstrate several compelling use cases of Delphi, including its incorporation as a component within an ensemble of AI systems. Finally, we computationally demonstrate the potential of Rawls’s prospect of hybrid approaches for reliable moral reasoning, inspiring future research in computational morality. Aligning artificial intelligence systems with human morality poses scientific, societal and ethical challenges. Delphi, an artificial intelligence system designed to predict human moral judgements based on John Rawls’s philosophical framework, is developed and tested, highlighting its potential for ethical applications and emphasizing the need to address its limitations and biases.","PeriodicalId":48533,"journal":{"name":"Nature Machine Intelligence","volume":"7 1","pages":"145-160"},"PeriodicalIF":18.8,"publicationDate":"2025-01-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.nature.com/articles/s42256-024-00969-6.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142968287","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 : 2025-01-06DOI: 10.1038/s42256-024-00960-1
Yonghan Yu, Ming Li
Peptide identification in mass spectrometry-based proteomics is crucial for understanding protein function and dynamics. Traditional database search methods, though widely used, rely on heuristic scoring functions, and statistical estimations must be introduced to achieve a higher identification rate. Here we introduce DeepSearch, a deep learning-based end-to-end database search method for tandem mass spectrometry. DeepSearch leverages a modified transformer-based encoder–decoder architecture under the contrastive learning framework. Unlike conventional methods, which rely on ion-to-ion matching, DeepSearch adopts a data-driven approach to score peptide–spectrum matches. DeepSearch can also profile variable post-translational modifications in a zero-shot manner. We show that DeepSearch’s scoring scheme expresses less bias and does not require any statistical estimation. We validate DeepSearch’s accuracy and robustness across various datasets, including those from species with diverse protein compositions and a modification-enriched dataset. DeepSearch sheds new light on database search methods in tandem mass spectrometry. Yu and Li present DeepSearch, a deep learning-based method for peptide identification in mass spectrometry, offering unbiased, data-driven scoring without statistical estimation. It accurately profiles post-translational modifications in a zero-shot manner.
{"title":"Towards highly sensitive deep learning-based end-to-end database search for tandem mass spectrometry","authors":"Yonghan Yu, Ming Li","doi":"10.1038/s42256-024-00960-1","DOIUrl":"10.1038/s42256-024-00960-1","url":null,"abstract":"Peptide identification in mass spectrometry-based proteomics is crucial for understanding protein function and dynamics. Traditional database search methods, though widely used, rely on heuristic scoring functions, and statistical estimations must be introduced to achieve a higher identification rate. Here we introduce DeepSearch, a deep learning-based end-to-end database search method for tandem mass spectrometry. DeepSearch leverages a modified transformer-based encoder–decoder architecture under the contrastive learning framework. Unlike conventional methods, which rely on ion-to-ion matching, DeepSearch adopts a data-driven approach to score peptide–spectrum matches. DeepSearch can also profile variable post-translational modifications in a zero-shot manner. We show that DeepSearch’s scoring scheme expresses less bias and does not require any statistical estimation. We validate DeepSearch’s accuracy and robustness across various datasets, including those from species with diverse protein compositions and a modification-enriched dataset. DeepSearch sheds new light on database search methods in tandem mass spectrometry. Yu and Li present DeepSearch, a deep learning-based method for peptide identification in mass spectrometry, offering unbiased, data-driven scoring without statistical estimation. It accurately profiles post-translational modifications in a zero-shot manner.","PeriodicalId":48533,"journal":{"name":"Nature Machine Intelligence","volume":"7 1","pages":"85-95"},"PeriodicalIF":18.8,"publicationDate":"2025-01-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142929493","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}
Machine learning (ML) techniques are gaining traction for materials image processing applications. In this context, Ziatdinov et al. developed AtomAI, a user-friendly and comprehensive Python library designed for a wide range of materials imaging tasks, including image segmentation, denoising, image generation, image-to-spectrum mapping (and vice versa) and subsequent atomistic modelling of image-resolved structures. Given its broad applicability, this report aims to reproduce key aspects of the authors’ original work, extend its capabilities to new materials datasets and enhance certain features to improve model performance. We have not only successfully replicated parts of the original study, but also developed improved ML models for multiple datasets across different image processing tasks. The AtomAI library was found to be easy to use and extensible for custom applications. We believe that AtomAI holds significant potential for the microscopy and spectroscopy communities, and further development—such as semi-automated image segmentation—could broaden its utility and impact. Vashishtha and colleagues test and reuse AtomAI, a machine learning framework developed for analysing microscopy data, for a range of materials characterization tasks.
{"title":"Reusability report: Deep learning-based analysis of images and spectroscopy data with AtomAI","authors":"Pragalbh Vashishtha, Hitesh Gupta Kattamuri, Nikhil Thawari, Murugaiyan Amirthalingam, Rohit Batra","doi":"10.1038/s42256-024-00958-9","DOIUrl":"10.1038/s42256-024-00958-9","url":null,"abstract":"Machine learning (ML) techniques are gaining traction for materials image processing applications. In this context, Ziatdinov et al. developed AtomAI, a user-friendly and comprehensive Python library designed for a wide range of materials imaging tasks, including image segmentation, denoising, image generation, image-to-spectrum mapping (and vice versa) and subsequent atomistic modelling of image-resolved structures. Given its broad applicability, this report aims to reproduce key aspects of the authors’ original work, extend its capabilities to new materials datasets and enhance certain features to improve model performance. We have not only successfully replicated parts of the original study, but also developed improved ML models for multiple datasets across different image processing tasks. The AtomAI library was found to be easy to use and extensible for custom applications. We believe that AtomAI holds significant potential for the microscopy and spectroscopy communities, and further development—such as semi-automated image segmentation—could broaden its utility and impact. Vashishtha and colleagues test and reuse AtomAI, a machine learning framework developed for analysing microscopy data, for a range of materials characterization tasks.","PeriodicalId":48533,"journal":{"name":"Nature Machine Intelligence","volume":"7 1","pages":"79-84"},"PeriodicalIF":18.8,"publicationDate":"2025-01-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142911796","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}
SARS-CoV-2 mutations accumulated during the COVID-19 pandemic, posing significant challenges for immune prevention. An optimistic perspective suggests that SARS-CoV-2 will become more tropic to humans with weaker virulence and stronger infectivity. However, tracing a quantified trajectory of this process remains difficult. Here we introduce an attentional recurrent network based on language embedding (ARNLE) framework to analyse the shift in SARS-CoV-2 host tropism towards humans. ARNLE incorporates a language model for self-supervised learning to capture the features of amino acid sequences, alongside a supervised bidirectional long-short-term-memory-based network to discern the relationship between mutations and host tropism among coronaviruses. We identified a shift in SARS-CoV-2 tropism from weak to strong, transitioning from an approximate Chiroptera coronavirus to a primate-tropic coronavirus. Delta variants were closer to other common primate coronaviruses than previous SARS-CoV-2 variants. A similar phenomenon was observed among the Omicron variants. We employed a Bayesian-based post hoc explanation method to analyse key mutations influencing the human tropism of SARS-CoV-2. ARNLE identified pivotal mutations in the spike proteins, including T478K, L452R, G142D and so on, as the top determinants of human tropism. Our findings suggest that language models like ARNLE will significantly facilitate the identification of potentially prevalent variants and provide important support for screening key mutations, aiding in timely update of vaccines to protect against future emerging SARS-CoV-2 variants. Liu et al. developed a framework called ARNLE to explore host tropism of SARS-CoV-2 and found a shift from weak to strong primate tropism. Key mutations involved in this shift can be analysed to advance research on emerging viruses.
{"title":"ARNLE model identifies prevalence potential of SARS-CoV-2 variants","authors":"Yuqi Liu, Jing Li, Peihan Li, Yehong Yang, Kaiying Wang, Jinhui Li, Lang Yang, Jiangfeng Liu, Leili Jia, Aiping Wu, Juntao Yang, Peng Li, Hongbin Song","doi":"10.1038/s42256-024-00919-2","DOIUrl":"10.1038/s42256-024-00919-2","url":null,"abstract":"SARS-CoV-2 mutations accumulated during the COVID-19 pandemic, posing significant challenges for immune prevention. An optimistic perspective suggests that SARS-CoV-2 will become more tropic to humans with weaker virulence and stronger infectivity. However, tracing a quantified trajectory of this process remains difficult. Here we introduce an attentional recurrent network based on language embedding (ARNLE) framework to analyse the shift in SARS-CoV-2 host tropism towards humans. ARNLE incorporates a language model for self-supervised learning to capture the features of amino acid sequences, alongside a supervised bidirectional long-short-term-memory-based network to discern the relationship between mutations and host tropism among coronaviruses. We identified a shift in SARS-CoV-2 tropism from weak to strong, transitioning from an approximate Chiroptera coronavirus to a primate-tropic coronavirus. Delta variants were closer to other common primate coronaviruses than previous SARS-CoV-2 variants. A similar phenomenon was observed among the Omicron variants. We employed a Bayesian-based post hoc explanation method to analyse key mutations influencing the human tropism of SARS-CoV-2. ARNLE identified pivotal mutations in the spike proteins, including T478K, L452R, G142D and so on, as the top determinants of human tropism. Our findings suggest that language models like ARNLE will significantly facilitate the identification of potentially prevalent variants and provide important support for screening key mutations, aiding in timely update of vaccines to protect against future emerging SARS-CoV-2 variants. Liu et al. developed a framework called ARNLE to explore host tropism of SARS-CoV-2 and found a shift from weak to strong primate tropism. Key mutations involved in this shift can be analysed to advance research on emerging viruses.","PeriodicalId":48533,"journal":{"name":"Nature Machine Intelligence","volume":"7 1","pages":"18-28"},"PeriodicalIF":18.8,"publicationDate":"2024-12-31","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142904754","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 : 2024-12-31DOI: 10.1038/s42256-024-00950-3
Ismael T. Freire, Adrián F. Amil, Paul F. M. J. Verschure
Deep reinforcement learning algorithms are known for their sample inefficiency, requiring extensive episodes to reach optimal performance. Episodic reinforcement learning algorithms aim to overcome this issue by using extended memory systems to leverage past experiences. However, these memory augmentations are often used as mere buffers, from which isolated events are resampled for offline learning (for example, replay). In this Article, we introduce Sequential Episodic Control (SEC), a hippocampal-inspired model that stores entire event sequences in their temporal order and employs a sequential bias in their retrieval to guide actions. We evaluate SEC across various benchmarks from the Animal-AI testbed, demonstrating its superior performance and sample efficiency compared to several state-of-the-art models, including Model-Free Episodic Control, Deep Q-Network and Episodic Reinforcement Learning with Associative Memory. Our experiments show that SEC achieves higher rewards and faster policy convergence in tasks requiring memory and decision-making. Additionally, we investigate the effects of memory constraints and forgetting mechanisms, revealing that prioritized forgetting enhances both performance and policy stability. Further, ablation studies demonstrate the critical role of the sequential memory component in SEC. Finally, we discuss how fast, sequential hippocampal-like episodic memory systems could support both habit formation and deliberation in artificial and biological systems. Previous studies have explored the integration of episodic memory into reinforcement learning and control. Inspired by hippocampal memory, Freire et al. develop a model that improves learning speed and stability by storing experiences as sequences, demonstrating resilience and efficiency under memory constraints.
{"title":"Sequential memory improves sample and memory efficiency in episodic control","authors":"Ismael T. Freire, Adrián F. Amil, Paul F. M. J. Verschure","doi":"10.1038/s42256-024-00950-3","DOIUrl":"10.1038/s42256-024-00950-3","url":null,"abstract":"Deep reinforcement learning algorithms are known for their sample inefficiency, requiring extensive episodes to reach optimal performance. Episodic reinforcement learning algorithms aim to overcome this issue by using extended memory systems to leverage past experiences. However, these memory augmentations are often used as mere buffers, from which isolated events are resampled for offline learning (for example, replay). In this Article, we introduce Sequential Episodic Control (SEC), a hippocampal-inspired model that stores entire event sequences in their temporal order and employs a sequential bias in their retrieval to guide actions. We evaluate SEC across various benchmarks from the Animal-AI testbed, demonstrating its superior performance and sample efficiency compared to several state-of-the-art models, including Model-Free Episodic Control, Deep Q-Network and Episodic Reinforcement Learning with Associative Memory. Our experiments show that SEC achieves higher rewards and faster policy convergence in tasks requiring memory and decision-making. Additionally, we investigate the effects of memory constraints and forgetting mechanisms, revealing that prioritized forgetting enhances both performance and policy stability. Further, ablation studies demonstrate the critical role of the sequential memory component in SEC. Finally, we discuss how fast, sequential hippocampal-like episodic memory systems could support both habit formation and deliberation in artificial and biological systems. Previous studies have explored the integration of episodic memory into reinforcement learning and control. Inspired by hippocampal memory, Freire et al. develop a model that improves learning speed and stability by storing experiences as sequences, demonstrating resilience and efficiency under memory constraints.","PeriodicalId":48533,"journal":{"name":"Nature Machine Intelligence","volume":"7 1","pages":"43-55"},"PeriodicalIF":18.8,"publicationDate":"2024-12-31","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142904749","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 : 2024-12-27DOI: 10.1038/s42256-024-00934-3
Till Richter, Mojtaba Bahrami, Yufan Xia, David S. Fischer, Fabian J. Theis
Self-supervised learning (SSL) has emerged as a powerful method for extracting meaningful representations from vast, unlabelled datasets, transforming computer vision and natural language processing. In single-cell genomics (SCG), representation learning offers insights into the complex biological data, especially with emerging foundation models. However, identifying scenarios in SCG where SSL outperforms traditional learning methods remains a nuanced challenge. Furthermore, selecting the most effective pretext tasks within the SSL framework for SCG is a critical yet unresolved question. Here we address this gap by adapting and benchmarking SSL methods in SCG, including masked autoencoders with multiple masking strategies and contrastive learning methods. Models trained on over 20 million cells were examined across multiple downstream tasks, including cell-type prediction, gene-expression reconstruction, cross-modality prediction and data integration. Our empirical analyses underscore the nuanced role of SSL, namely, in transfer learning scenarios leveraging auxiliary data or analysing unseen datasets. Masked autoencoders excel over contrastive methods in SCG, diverging from computer vision trends. Moreover, our findings reveal the notable capabilities of SSL in zero-shot settings and its potential in cross-modality prediction and data integration. In summary, we study SSL methods in SCG on fully connected networks and benchmark their utility across key representation learning scenarios. Self-supervised learning techniques are powerful assets for enabling deep insights into complex, unlabelled single-cell genomic data. Richter et al. here benchmark the applicability of self-supervised architectures into key downstream representation learning scenarios.
{"title":"Delineating the effective use of self-supervised learning in single-cell genomics","authors":"Till Richter, Mojtaba Bahrami, Yufan Xia, David S. Fischer, Fabian J. Theis","doi":"10.1038/s42256-024-00934-3","DOIUrl":"10.1038/s42256-024-00934-3","url":null,"abstract":"Self-supervised learning (SSL) has emerged as a powerful method for extracting meaningful representations from vast, unlabelled datasets, transforming computer vision and natural language processing. In single-cell genomics (SCG), representation learning offers insights into the complex biological data, especially with emerging foundation models. However, identifying scenarios in SCG where SSL outperforms traditional learning methods remains a nuanced challenge. Furthermore, selecting the most effective pretext tasks within the SSL framework for SCG is a critical yet unresolved question. Here we address this gap by adapting and benchmarking SSL methods in SCG, including masked autoencoders with multiple masking strategies and contrastive learning methods. Models trained on over 20 million cells were examined across multiple downstream tasks, including cell-type prediction, gene-expression reconstruction, cross-modality prediction and data integration. Our empirical analyses underscore the nuanced role of SSL, namely, in transfer learning scenarios leveraging auxiliary data or analysing unseen datasets. Masked autoencoders excel over contrastive methods in SCG, diverging from computer vision trends. Moreover, our findings reveal the notable capabilities of SSL in zero-shot settings and its potential in cross-modality prediction and data integration. In summary, we study SSL methods in SCG on fully connected networks and benchmark their utility across key representation learning scenarios. Self-supervised learning techniques are powerful assets for enabling deep insights into complex, unlabelled single-cell genomic data. Richter et al. here benchmark the applicability of self-supervised architectures into key downstream representation learning scenarios.","PeriodicalId":48533,"journal":{"name":"Nature Machine Intelligence","volume":"7 1","pages":"68-78"},"PeriodicalIF":18.8,"publicationDate":"2024-12-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.nature.com/articles/s42256-024-00934-3.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142888262","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 : 2024-12-18DOI: 10.1038/s42256-024-00968-7
Clear descriptions of intelligence in both living organisms and machines are essential to avoid confusion, sharpen thinking and guide interdisciplinary research. A Comment in this issue encourages researchers to answer key questions to improve clarity on the terms they use.
{"title":"Seeking clarity rather than strong opinions on intelligence","authors":"","doi":"10.1038/s42256-024-00968-7","DOIUrl":"10.1038/s42256-024-00968-7","url":null,"abstract":"Clear descriptions of intelligence in both living organisms and machines are essential to avoid confusion, sharpen thinking and guide interdisciplinary research. A Comment in this issue encourages researchers to answer key questions to improve clarity on the terms they use.","PeriodicalId":48533,"journal":{"name":"Nature Machine Intelligence","volume":"6 12","pages":"1408-1408"},"PeriodicalIF":18.8,"publicationDate":"2024-12-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.nature.com/articles/s42256-024-00968-7.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142841571","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 : 2024-12-18DOI: 10.1038/s42256-024-00967-8
Researchers urgently need more guidance to help them identify and mitigate potential risks when designing projects that involve AI developments.
研究人员迫切需要更多的指导，以帮助他们在设计涉及人工智能开发的项目时识别和减轻潜在风险。
{"title":"Strategies needed to counter potential AI misuse","authors":"","doi":"10.1038/s42256-024-00967-8","DOIUrl":"10.1038/s42256-024-00967-8","url":null,"abstract":"Researchers urgently need more guidance to help them identify and mitigate potential risks when designing projects that involve AI developments.","PeriodicalId":48533,"journal":{"name":"Nature Machine Intelligence","volume":"6 12","pages":"1407-1407"},"PeriodicalIF":18.8,"publicationDate":"2024-12-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.nature.com/articles/s42256-024-00967-8.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142841570","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 : 2024-12-18DOI: 10.1038/s42256-024-00935-2
D. S. Matthews, M. A. Spence, A. C. Mater, J. Nichols, S. B. Pulsford, M. Sandhu, J. A. Kaczmarski, C. M. Miton, N. Tokuriki, C. J. Jackson
Protein language models (PLMs) convert amino acid sequences into the numerical representations required to train machine learning models. Many PLMs are large (>600 million parameters) and trained on a broad span of protein sequence space. However, these models have limitations in terms of predictive accuracy and computational cost. Here we use multiplexed ancestral sequence reconstruction to generate small but focused functional protein sequence datasets for PLM training. Compared to large PLMs, this local ancestral sequence embedding produces representations with higher predictive accuracy. We show that due to the evolutionary nature of the ancestral sequence reconstruction data, local ancestral sequence embedding produces smoother fitness landscapes, in which protein variants that are closer in fitness value become numerically closer in representation space. This work contributes to the implementation of machine learning-based protein design in real-world settings, where data are sparse and computational resources are limited. Matthews et al. present a protein sequence embedding based on data from ancestral sequences that allows machine learning to be used for tasks where training data are scarce or expensive.
{"title":"Leveraging ancestral sequence reconstruction for protein representation learning","authors":"D. S. Matthews, M. A. Spence, A. C. Mater, J. Nichols, S. B. Pulsford, M. Sandhu, J. A. Kaczmarski, C. M. Miton, N. Tokuriki, C. J. Jackson","doi":"10.1038/s42256-024-00935-2","DOIUrl":"10.1038/s42256-024-00935-2","url":null,"abstract":"Protein language models (PLMs) convert amino acid sequences into the numerical representations required to train machine learning models. Many PLMs are large (>600 million parameters) and trained on a broad span of protein sequence space. However, these models have limitations in terms of predictive accuracy and computational cost. Here we use multiplexed ancestral sequence reconstruction to generate small but focused functional protein sequence datasets for PLM training. Compared to large PLMs, this local ancestral sequence embedding produces representations with higher predictive accuracy. We show that due to the evolutionary nature of the ancestral sequence reconstruction data, local ancestral sequence embedding produces smoother fitness landscapes, in which protein variants that are closer in fitness value become numerically closer in representation space. This work contributes to the implementation of machine learning-based protein design in real-world settings, where data are sparse and computational resources are limited. Matthews et al. present a protein sequence embedding based on data from ancestral sequences that allows machine learning to be used for tasks where training data are scarce or expensive.","PeriodicalId":48533,"journal":{"name":"Nature Machine Intelligence","volume":"6 12","pages":"1542-1555"},"PeriodicalIF":18.8,"publicationDate":"2024-12-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142841572","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 : 2024-12-16DOI: 10.1038/s42256-024-00951-2
Roy Moyal, Nabil Imam, Thomas A. Cleland
{"title":"Reply to: Limitations in odour recognition and generalization in a neuromorphic olfactory circuit","authors":"Roy Moyal, Nabil Imam, Thomas A. Cleland","doi":"10.1038/s42256-024-00951-2","DOIUrl":"10.1038/s42256-024-00951-2","url":null,"abstract":"","PeriodicalId":48533,"journal":{"name":"Nature Machine Intelligence","volume":"6 12","pages":"1454-1456"},"PeriodicalIF":18.8,"publicationDate":"2024-12-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142825246","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}
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：