Protein glycosylation, a post-translational modification of proteins by glycans, plays an important role in numerous physiological and pathological cellular functions. Glycoproteomics, the study of protein glycosylation on a proteome-wide scale, utilizes liquid chromatography coupled with tandem mass spectrometry (MS/MS) to get combinational information on glycosylation site, glycosylation level and glycan structure. However, current database searching methods for glycoproteomics often struggle with glycan structure determination due to the limited occurrence of structure-determining ions. Although spectral searching methods can leverage fragment intensity to facilitate the structure identification of glycopeptides, their application is hindered by difficulties in spectral library construction. In this work, we present DeepGP, a hybrid deep learning framework based on transformer and graph neural networks, for the prediction of MS/MS spectra and retention time of glycopeptides. Two graph neural network modules are employed to capture the branched glycan structure and predict glycan ion intensity, respectively. Additionally, a pretraining strategy is implemented to alleviate the insufficiency of glycoproteomics data. Testing on multiple biological datasets, DeepGP accurately predicts MS/MS spectra and retention time of glycopeptides, closely aligning with the experimental results. Comprehensive benchmarking of DeepGP on synthetic and biological datasets validates its effectiveness in distinguishing similar glycans. Based on various decoy methods, DeepGP in combination with database searching can increase glycopeptide detection sensitivity. We anticipate that DeepGP can inspire extensive future work in glycoproteomics. Glycosylation, a prevalent type of post-translational modification of proteins by glycan molecules, plays a major role in the proteome. Zong et al. present DeepGP, a hybrid deep learning framework based on transformer and graph neural network architectures that accurately predicts tandem mass spectra and retention times of glycopeptides, providing information on glycosylation and glycan structure.
{"title":"Deep learning prediction of glycopeptide tandem mass spectra powers glycoproteomics","authors":"Yu Zong, Yuxin Wang, Xipeng Qiu, Xuanjing Huang, Liang Qiao","doi":"10.1038/s42256-024-00875-x","DOIUrl":"10.1038/s42256-024-00875-x","url":null,"abstract":"Protein glycosylation, a post-translational modification of proteins by glycans, plays an important role in numerous physiological and pathological cellular functions. Glycoproteomics, the study of protein glycosylation on a proteome-wide scale, utilizes liquid chromatography coupled with tandem mass spectrometry (MS/MS) to get combinational information on glycosylation site, glycosylation level and glycan structure. However, current database searching methods for glycoproteomics often struggle with glycan structure determination due to the limited occurrence of structure-determining ions. Although spectral searching methods can leverage fragment intensity to facilitate the structure identification of glycopeptides, their application is hindered by difficulties in spectral library construction. In this work, we present DeepGP, a hybrid deep learning framework based on transformer and graph neural networks, for the prediction of MS/MS spectra and retention time of glycopeptides. Two graph neural network modules are employed to capture the branched glycan structure and predict glycan ion intensity, respectively. Additionally, a pretraining strategy is implemented to alleviate the insufficiency of glycoproteomics data. Testing on multiple biological datasets, DeepGP accurately predicts MS/MS spectra and retention time of glycopeptides, closely aligning with the experimental results. Comprehensive benchmarking of DeepGP on synthetic and biological datasets validates its effectiveness in distinguishing similar glycans. Based on various decoy methods, DeepGP in combination with database searching can increase glycopeptide detection sensitivity. We anticipate that DeepGP can inspire extensive future work in glycoproteomics. Glycosylation, a prevalent type of post-translational modification of proteins by glycan molecules, plays a major role in the proteome. Zong et al. present DeepGP, a hybrid deep learning framework based on transformer and graph neural network architectures that accurately predicts tandem mass spectra and retention times of glycopeptides, providing information on glycosylation and glycan structure.","PeriodicalId":48533,"journal":{"name":"Nature Machine Intelligence","volume":"6 8","pages":"950-961"},"PeriodicalIF":18.8,"publicationDate":"2024-07-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141794616","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 : 2024-07-29DOI: 10.1038/s42256-024-00877-9
Silvia Milano, Sven Nyholm
{"title":"Advanced AI assistants that act on our behalf may not be ethically or legally feasible","authors":"Silvia Milano, Sven Nyholm","doi":"10.1038/s42256-024-00877-9","DOIUrl":"10.1038/s42256-024-00877-9","url":null,"abstract":"","PeriodicalId":48533,"journal":{"name":"Nature Machine Intelligence","volume":"6 8","pages":"846-847"},"PeriodicalIF":18.8,"publicationDate":"2024-07-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141791118","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 : 2024-07-24DOI: 10.1038/s42256-024-00880-0
Medical research is one of the most impactful areas for machine learning applications, but access to large and diverse health datasets is needed for models to be useful. Winning trust from patients by demonstrating that data are handled securely and effectively is key.
{"title":"A question of trust for AI research in medicine","authors":"","doi":"10.1038/s42256-024-00880-0","DOIUrl":"10.1038/s42256-024-00880-0","url":null,"abstract":"Medical research is one of the most impactful areas for machine learning applications, but access to large and diverse health datasets is needed for models to be useful. Winning trust from patients by demonstrating that data are handled securely and effectively is key.","PeriodicalId":48533,"journal":{"name":"Nature Machine Intelligence","volume":"6 7","pages":"739-739"},"PeriodicalIF":18.8,"publicationDate":"2024-07-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.nature.com/articles/s42256-024-00880-0.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141764060","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 : 2024-07-23DOI: 10.1038/s42256-024-00872-0
Melissa Sanabria, Jonas Hirsch, Pierre M. Joubert, Anna R. Poetsch
Deep-learning models that learn a sense of language on DNA have achieved a high level of performance on genome biological tasks. Genome sequences follow rules similar to natural language but are distinct in the absence of a concept of words. We established byte-pair encoding on the human genome and trained a foundation language model called GROVER (Genome Rules Obtained Via Extracted Representations) with the vocabulary selected via a custom task, next-k-mer prediction. The defined dictionary of tokens in the human genome carries best the information content for GROVER. Analysing learned representations, we observed that trained token embeddings primarily encode information related to frequency, sequence content and length. Some tokens are primarily localized in repeats, whereas the majority widely distribute over the genome. GROVER also learns context and lexical ambiguity. Average trained embeddings of genomic regions relate to functional genomics annotation and thus indicate learning of these structures purely from the contextual relationships of tokens. This highlights the extent of information content encoded by the sequence that can be grasped by GROVER. On fine-tuning tasks addressing genome biology with questions of genome element identification and protein–DNA binding, GROVER exceeds other models’ performance. GROVER learns sequence context, a sense for structure and language rules. Extracting this knowledge can be used to compose a grammar book for the code of life. Genomes can be modelled with language approaches by treating nucleotide bases A, C, G and T like text, but there is no natural concept of what the words would be and whether there is even a ‘language’ to be learned this way. Sanabria et al. have developed a language model called GROVER that learns with a ‘vocabulary’ of genome sequences with byte-pair encoding, a method from text compression, and shows good performance on genome biological tasks.
{"title":"DNA language model GROVER learns sequence context in the human genome","authors":"Melissa Sanabria, Jonas Hirsch, Pierre M. Joubert, Anna R. Poetsch","doi":"10.1038/s42256-024-00872-0","DOIUrl":"10.1038/s42256-024-00872-0","url":null,"abstract":"Deep-learning models that learn a sense of language on DNA have achieved a high level of performance on genome biological tasks. Genome sequences follow rules similar to natural language but are distinct in the absence of a concept of words. We established byte-pair encoding on the human genome and trained a foundation language model called GROVER (Genome Rules Obtained Via Extracted Representations) with the vocabulary selected via a custom task, next-k-mer prediction. The defined dictionary of tokens in the human genome carries best the information content for GROVER. Analysing learned representations, we observed that trained token embeddings primarily encode information related to frequency, sequence content and length. Some tokens are primarily localized in repeats, whereas the majority widely distribute over the genome. GROVER also learns context and lexical ambiguity. Average trained embeddings of genomic regions relate to functional genomics annotation and thus indicate learning of these structures purely from the contextual relationships of tokens. This highlights the extent of information content encoded by the sequence that can be grasped by GROVER. On fine-tuning tasks addressing genome biology with questions of genome element identification and protein–DNA binding, GROVER exceeds other models’ performance. GROVER learns sequence context, a sense for structure and language rules. Extracting this knowledge can be used to compose a grammar book for the code of life. Genomes can be modelled with language approaches by treating nucleotide bases A, C, G and T like text, but there is no natural concept of what the words would be and whether there is even a ‘language’ to be learned this way. Sanabria et al. have developed a language model called GROVER that learns with a ‘vocabulary’ of genome sequences with byte-pair encoding, a method from text compression, and shows good performance on genome biological tasks.","PeriodicalId":48533,"journal":{"name":"Nature Machine Intelligence","volume":"6 8","pages":"911-923"},"PeriodicalIF":18.8,"publicationDate":"2024-07-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.nature.com/articles/s42256-024-00872-0.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141750334","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 : 2024-07-22DOI: 10.1038/s42256-024-00867-x
Yoo-Geun Ham, Yong-Sik Joo, Jeong-Hwan Kim, Jeong-Gil Lee
The oceanic data assimilation (DA) system has been developed to optimally combine numerical-model predictions with actual measurements from the ocean to create the best estimates of current ocean conditions and their uncertainties, improving our ability to forecast and understand the global climate variations. We developed DeepDA, a global oceanic DA system using deep learning, by integrating a partial convolutional neural network and a generative adversarial network. Partial convolution serves as an observation operator, mapping irregular observational data onto gridded fields, while generative adversarial network incorporates observational information from previous time frames. Our observing system simulation experiments, using simulated observations for the DA, revealed that DeepDA markedly reduces analysis error of the oceanic temperature, outperforming both background and observed values. DeepDA’s real-case global temperature reanalysis spanning from 1981 to 2020 accurately reconstructs observed global climatological temperature fields, along with their seasonal cycles, major oceanic temperature variabilities and global warming trend. Developed solely with a long-term control simulation, DeepDA lowers technical hurdles in creating global ocean reanalysis datasets using multiple numerical models’ physical constraints, thereby diminishing systematic uncertainties in estimating global oceanic states over decades with these models. Data assimilation (DA) techniques are commonly used to assess global Earth system variability but require considerable computational resources and struggle to handle sparse observational data. Ham and colleagues introduce a partial convolution and generative adversarial network-based global oceanic DA system and successfully reconstruct the observed global temperature in a real case study with smaller computational costs than traditional DA systems.
{"title":"Partial-convolution-implemented generative adversarial network for global oceanic data assimilation","authors":"Yoo-Geun Ham, Yong-Sik Joo, Jeong-Hwan Kim, Jeong-Gil Lee","doi":"10.1038/s42256-024-00867-x","DOIUrl":"10.1038/s42256-024-00867-x","url":null,"abstract":"The oceanic data assimilation (DA) system has been developed to optimally combine numerical-model predictions with actual measurements from the ocean to create the best estimates of current ocean conditions and their uncertainties, improving our ability to forecast and understand the global climate variations. We developed DeepDA, a global oceanic DA system using deep learning, by integrating a partial convolutional neural network and a generative adversarial network. Partial convolution serves as an observation operator, mapping irregular observational data onto gridded fields, while generative adversarial network incorporates observational information from previous time frames. Our observing system simulation experiments, using simulated observations for the DA, revealed that DeepDA markedly reduces analysis error of the oceanic temperature, outperforming both background and observed values. DeepDA’s real-case global temperature reanalysis spanning from 1981 to 2020 accurately reconstructs observed global climatological temperature fields, along with their seasonal cycles, major oceanic temperature variabilities and global warming trend. Developed solely with a long-term control simulation, DeepDA lowers technical hurdles in creating global ocean reanalysis datasets using multiple numerical models’ physical constraints, thereby diminishing systematic uncertainties in estimating global oceanic states over decades with these models. Data assimilation (DA) techniques are commonly used to assess global Earth system variability but require considerable computational resources and struggle to handle sparse observational data. Ham and colleagues introduce a partial convolution and generative adversarial network-based global oceanic DA system and successfully reconstruct the observed global temperature in a real case study with smaller computational costs than traditional DA systems.","PeriodicalId":48533,"journal":{"name":"Nature Machine Intelligence","volume":"6 7","pages":"834-843"},"PeriodicalIF":18.8,"publicationDate":"2024-07-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141736940","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}
The complex diagnostic criteria for gliomas pose great challenges for making accurate diagnoses with computational pathology methods. There are no in-depth analyses of the accuracy, reliability and auxiliary capability of present approaches from a clinical perspective. Previous studies have overlooked the exploration of molecular and morphological correlations. To overcome these limitations, we propose ROAM, a multiple-instance learning model based on large regions of interest and a pyramid transformer. ROAM enlarges regions of interest to facilitate the consideration of tissue contexts. It utilizes the pyramid transformer to model both intrascale and interscale correlations of morphological features and leverages class-specific multiple-instance learning based on attention to extract slide-level visual representations that can be used to diagnose gliomas. Through comprehensive experiments on both in-house and external glioma datasets, we demonstrate that ROAM can automatically capture key morphological features consistent with the experience of pathologists and thus provide accurate, reliable and adaptable clinical-grade diagnoses of gliomas. Moreover, ROAM has clinical value for auxiliary diagnoses and could pave the way for the study of molecular and morphological correlations. ROAM, based on large regions of interest and a pyramid transformer, can automatically capture key morphological features consistent with the experience of pathologists to provide accurate, reliable and adaptable clinical-grade diagnoses of gliomas while advancing the discovery of molecular and morphological markers related to glioma diagnosis.
{"title":"A transformer-based weakly supervised computational pathology method for clinical-grade diagnosis and molecular marker discovery of gliomas","authors":"Rui Jiang, Xiaoxu Yin, Pengshuai Yang, Lingchao Cheng, Juan Hu, Jiao Yang, Ying Wang, Xiaodan Fu, Li Shang, Liling Li, Wei Lin, Huan Zhou, Fufeng Chen, Xuegong Zhang, Zhongliang Hu, Hairong Lv","doi":"10.1038/s42256-024-00868-w","DOIUrl":"10.1038/s42256-024-00868-w","url":null,"abstract":"The complex diagnostic criteria for gliomas pose great challenges for making accurate diagnoses with computational pathology methods. There are no in-depth analyses of the accuracy, reliability and auxiliary capability of present approaches from a clinical perspective. Previous studies have overlooked the exploration of molecular and morphological correlations. To overcome these limitations, we propose ROAM, a multiple-instance learning model based on large regions of interest and a pyramid transformer. ROAM enlarges regions of interest to facilitate the consideration of tissue contexts. It utilizes the pyramid transformer to model both intrascale and interscale correlations of morphological features and leverages class-specific multiple-instance learning based on attention to extract slide-level visual representations that can be used to diagnose gliomas. Through comprehensive experiments on both in-house and external glioma datasets, we demonstrate that ROAM can automatically capture key morphological features consistent with the experience of pathologists and thus provide accurate, reliable and adaptable clinical-grade diagnoses of gliomas. Moreover, ROAM has clinical value for auxiliary diagnoses and could pave the way for the study of molecular and morphological correlations. ROAM, based on large regions of interest and a pyramid transformer, can automatically capture key morphological features consistent with the experience of pathologists to provide accurate, reliable and adaptable clinical-grade diagnoses of gliomas while advancing the discovery of molecular and morphological markers related to glioma diagnosis.","PeriodicalId":48533,"journal":{"name":"Nature Machine Intelligence","volume":"6 8","pages":"876-891"},"PeriodicalIF":18.8,"publicationDate":"2024-07-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141725778","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 : 2024-07-18DOI: 10.1038/s42256-024-00863-1
James Gornet, Matt Thomson
Humans construct internal cognitive maps of their environment directly from sensory inputs without access to a system of explicit coordinates or distance measurements. Although machine learning algorithms like simultaneous localization and mapping utilize specialized inference procedures to identify visual features and construct spatial maps from visual and odometry data, the general nature of cognitive maps in the brain suggests a unified mapping algorithmic strategy that can generalize to auditory, tactile and linguistic inputs. Here we demonstrate that predictive coding provides a natural and versatile neural network algorithm for constructing spatial maps using sensory data. We introduce a framework in which an agent navigates a virtual environment while engaging in visual predictive coding using a self-attention-equipped convolutional neural network. While learning a next-image prediction task, the agent automatically constructs an internal representation of the environment that quantitatively reflects spatial distances. The internal map enables the agent to pinpoint its location relative to landmarks using only visual information.The predictive coding network generates a vectorized encoding of the environment that supports vector navigation, where individual latent space units delineate localized, overlapping neighbourhoods in the environment. Broadly, our work introduces predictive coding as a unified algorithmic framework for constructing cognitive maps that can naturally extend to the mapping of auditory, sensorimotor and linguistic inputs. Constructing spatial maps from sensory inputs is challenging in both neuroscience and artificial intelligence. Gornet and Thomson show that as an agent navigates an environment, a self-attention neural network using predictive coding can recover the environment’s map in its latent space.
{"title":"Automated construction of cognitive maps with visual predictive coding","authors":"James Gornet, Matt Thomson","doi":"10.1038/s42256-024-00863-1","DOIUrl":"10.1038/s42256-024-00863-1","url":null,"abstract":"Humans construct internal cognitive maps of their environment directly from sensory inputs without access to a system of explicit coordinates or distance measurements. Although machine learning algorithms like simultaneous localization and mapping utilize specialized inference procedures to identify visual features and construct spatial maps from visual and odometry data, the general nature of cognitive maps in the brain suggests a unified mapping algorithmic strategy that can generalize to auditory, tactile and linguistic inputs. Here we demonstrate that predictive coding provides a natural and versatile neural network algorithm for constructing spatial maps using sensory data. We introduce a framework in which an agent navigates a virtual environment while engaging in visual predictive coding using a self-attention-equipped convolutional neural network. While learning a next-image prediction task, the agent automatically constructs an internal representation of the environment that quantitatively reflects spatial distances. The internal map enables the agent to pinpoint its location relative to landmarks using only visual information.The predictive coding network generates a vectorized encoding of the environment that supports vector navigation, where individual latent space units delineate localized, overlapping neighbourhoods in the environment. Broadly, our work introduces predictive coding as a unified algorithmic framework for constructing cognitive maps that can naturally extend to the mapping of auditory, sensorimotor and linguistic inputs. Constructing spatial maps from sensory inputs is challenging in both neuroscience and artificial intelligence. Gornet and Thomson show that as an agent navigates an environment, a self-attention neural network using predictive coding can recover the environment’s map in its latent space.","PeriodicalId":48533,"journal":{"name":"Nature Machine Intelligence","volume":"6 7","pages":"820-833"},"PeriodicalIF":18.8,"publicationDate":"2024-07-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.nature.com/articles/s42256-024-00863-1.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141725785","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 : 2024-07-17DOI: 10.1038/s42256-024-00869-9
Robert Baines, Dylan Shah, Jeremy Marvel, Jennifer Case, Andrew Spielberg
{"title":"The need for reproducible research in soft robotics","authors":"Robert Baines, Dylan Shah, Jeremy Marvel, Jennifer Case, Andrew Spielberg","doi":"10.1038/s42256-024-00869-9","DOIUrl":"10.1038/s42256-024-00869-9","url":null,"abstract":"","PeriodicalId":48533,"journal":{"name":"Nature Machine Intelligence","volume":"6 7","pages":"740-741"},"PeriodicalIF":18.8,"publicationDate":"2024-07-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141631382","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 : 2024-07-15DOI: 10.1038/s42256-024-00864-0
Petru-Daniel Tudosiu, Walter H. L. Pinaya, Pedro Ferreira Da Costa, Jessica Dafflon, Ashay Patel, Pedro Borges, Virginia Fernandez, Mark S. Graham, Robert J. Gray, Parashkev Nachev, Sebastien Ourselin, M. Jorge Cardoso
Medical imaging research is often limited by data scarcity and availability. Governance, privacy concerns and the cost of acquisition all restrict access to medical imaging data, which, compounded by the data-hungry nature of deep learning algorithms, limits progress in the field of healthcare AI. Generative models have recently been used to synthesize photorealistic natural images, presenting a potential solution to the data scarcity problem. But are current generative models synthesizing morphologically correct samples? In this work we present a three-dimensional generative model of the human brain that is trained at the necessary scale to generate diverse, realistic-looking, high-resolution and morphologically preserving samples and conditioned on patient characteristics (for example, age and pathology). We show that the synthetic samples generated by the model preserve biological and disease phenotypes and are realistic enough to permit use downstream in well-established image analysis tools. While the proposed model has broad future applicability, such as anomaly detection and learning under limited data, its generative capabilities can be used to directly mitigate data scarcity, limited data availability and algorithmic fairness. Medical imaging research is limited by data availability. To address this challenge, Tudosiu and colleagues develop a 3D generative model of the human brain that can generate high-resolution morphologically correct brains conditioned on patient characteristics.
{"title":"Realistic morphology-preserving generative modelling of the brain","authors":"Petru-Daniel Tudosiu, Walter H. L. Pinaya, Pedro Ferreira Da Costa, Jessica Dafflon, Ashay Patel, Pedro Borges, Virginia Fernandez, Mark S. Graham, Robert J. Gray, Parashkev Nachev, Sebastien Ourselin, M. Jorge Cardoso","doi":"10.1038/s42256-024-00864-0","DOIUrl":"10.1038/s42256-024-00864-0","url":null,"abstract":"Medical imaging research is often limited by data scarcity and availability. Governance, privacy concerns and the cost of acquisition all restrict access to medical imaging data, which, compounded by the data-hungry nature of deep learning algorithms, limits progress in the field of healthcare AI. Generative models have recently been used to synthesize photorealistic natural images, presenting a potential solution to the data scarcity problem. But are current generative models synthesizing morphologically correct samples? In this work we present a three-dimensional generative model of the human brain that is trained at the necessary scale to generate diverse, realistic-looking, high-resolution and morphologically preserving samples and conditioned on patient characteristics (for example, age and pathology). We show that the synthetic samples generated by the model preserve biological and disease phenotypes and are realistic enough to permit use downstream in well-established image analysis tools. While the proposed model has broad future applicability, such as anomaly detection and learning under limited data, its generative capabilities can be used to directly mitigate data scarcity, limited data availability and algorithmic fairness. Medical imaging research is limited by data availability. To address this challenge, Tudosiu and colleagues develop a 3D generative model of the human brain that can generate high-resolution morphologically correct brains conditioned on patient characteristics.","PeriodicalId":48533,"journal":{"name":"Nature Machine Intelligence","volume":"6 7","pages":"811-819"},"PeriodicalIF":18.8,"publicationDate":"2024-07-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.nature.com/articles/s42256-024-00864-0.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141624661","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 : 2024-07-12DOI: 10.1038/s42256-024-00870-2
Yue Huang, Chengguang Zhu, Xiaokang Yang, Manhua Liu
The elucidation of three-dimensional (3D) structures is crucial for unravelling the protein function and illuminating mechanisms in structural biology. Cryogenic electron microscopy (cryo-EM) single-particle analysis provides direct measurements to determine the structures of macromolecules. However, the main challenge is reconstructing high-resolution 3D structures from extremely noisy and randomly oriented two-dimensional projection images. Most existing methods involve the optimization of multiple two-dimensional slices in the Fourier domain but ignore the anisotropy among these slices, thereby limiting the reconstruction of high-frequency structures. In this paper, we propose a cryo-EM neural field reconstruction network using 3D spatial-domain optimization that learns a directional isotropic representation of the cryo-EM structure by mapping the spatial coordinates to the corresponding density values. We qualitatively and quantitatively evaluate the cryo-EM neural field reconstruction network on four datasets. The cryo-EM neural field reconstruction network improves the directional isotropy and 3D density resolution beyond the limits of existing algorithms in homogeneous reconstruction and resolves the missing elements of SARS-CoV-2 in heterogeneous reconstruction. Elucidating three-dimensional structures is crucial for unravelling the macromolecule function in structural biology. This study presents a cryogenic electron microscopy neural field reconstruction network using real-space optimization, enhancing the resolution in cryogenic electron microscopy reconstruction.
{"title":"High-resolution real-space reconstruction of cryo-EM structures using a neural field network","authors":"Yue Huang, Chengguang Zhu, Xiaokang Yang, Manhua Liu","doi":"10.1038/s42256-024-00870-2","DOIUrl":"10.1038/s42256-024-00870-2","url":null,"abstract":"The elucidation of three-dimensional (3D) structures is crucial for unravelling the protein function and illuminating mechanisms in structural biology. Cryogenic electron microscopy (cryo-EM) single-particle analysis provides direct measurements to determine the structures of macromolecules. However, the main challenge is reconstructing high-resolution 3D structures from extremely noisy and randomly oriented two-dimensional projection images. Most existing methods involve the optimization of multiple two-dimensional slices in the Fourier domain but ignore the anisotropy among these slices, thereby limiting the reconstruction of high-frequency structures. In this paper, we propose a cryo-EM neural field reconstruction network using 3D spatial-domain optimization that learns a directional isotropic representation of the cryo-EM structure by mapping the spatial coordinates to the corresponding density values. We qualitatively and quantitatively evaluate the cryo-EM neural field reconstruction network on four datasets. The cryo-EM neural field reconstruction network improves the directional isotropy and 3D density resolution beyond the limits of existing algorithms in homogeneous reconstruction and resolves the missing elements of SARS-CoV-2 in heterogeneous reconstruction. Elucidating three-dimensional structures is crucial for unravelling the macromolecule function in structural biology. This study presents a cryogenic electron microscopy neural field reconstruction network using real-space optimization, enhancing the resolution in cryogenic electron microscopy reconstruction.","PeriodicalId":48533,"journal":{"name":"Nature Machine Intelligence","volume":"6 8","pages":"892-903"},"PeriodicalIF":18.8,"publicationDate":"2024-07-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141597695","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}