Luca Tomasetti, K. Engan, M. Khanmohammadi, K. D. Kurz
More than 13 million people suffer from ischemic cerebral stroke worldwide each year. Thrombolytic treatment can reduce brain damage but has a narrow treatment window. Computed Tomography Perfusion imaging is a commonly used primary assessment tool for stroke patients, and typically the radiologists will evaluate resulting parametric maps to estimate the affected areas, dead tissue (core), and the surrounding tissue at risk (penumbra), to decide further treatments. Different work has been reported, suggesting thresholds, and semi-automated methods, and in later years deep neural networks, for segmenting infarction areas based on the parametric maps. However, there is no consensus in terms of which thresholds to use, or how to combine the information from the parametric maps, and the presented methods all have limitations in terms of both accuracy and reproducibility. We propose a fully automated convolutional neural network based segmentation method that uses the full four-dimensional computed tomography perfusion dataset as input, rather than the pre-filtered parametric maps. The suggested network is tested on an available dataset as a proof-of-concept, with very encouraging results. Cross-validated results show averaged Dice score of 0.78 and 0.53, and an area under the receiver operating characteristic curve of 0.97 and 0.94 for penumbra and core respectively.
{"title":"CNN Based Segmentation of Infarcted Regions in Acute Cerebral Stroke Patients From Computed Tomography Perfusion Imaging","authors":"Luca Tomasetti, K. Engan, M. Khanmohammadi, K. D. Kurz","doi":"10.1145/3388440.3412470","DOIUrl":"https://doi.org/10.1145/3388440.3412470","url":null,"abstract":"More than 13 million people suffer from ischemic cerebral stroke worldwide each year. Thrombolytic treatment can reduce brain damage but has a narrow treatment window. Computed Tomography Perfusion imaging is a commonly used primary assessment tool for stroke patients, and typically the radiologists will evaluate resulting parametric maps to estimate the affected areas, dead tissue (core), and the surrounding tissue at risk (penumbra), to decide further treatments. Different work has been reported, suggesting thresholds, and semi-automated methods, and in later years deep neural networks, for segmenting infarction areas based on the parametric maps. However, there is no consensus in terms of which thresholds to use, or how to combine the information from the parametric maps, and the presented methods all have limitations in terms of both accuracy and reproducibility. We propose a fully automated convolutional neural network based segmentation method that uses the full four-dimensional computed tomography perfusion dataset as input, rather than the pre-filtered parametric maps. The suggested network is tested on an available dataset as a proof-of-concept, with very encouraging results. Cross-validated results show averaged Dice score of 0.78 and 0.53, and an area under the receiver operating characteristic curve of 0.97 and 0.94 for penumbra and core respectively.","PeriodicalId":411338,"journal":{"name":"Proceedings of the 11th ACM International Conference on Bioinformatics, Computational Biology and Health Informatics","volume":"12 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-09-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"117132112","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
The COVID-19 pandemic has infected millions of people around the world, spreading rapidly and causing a flood of patients that risk overwhelming clinical facilities. Whether in urban or rural areas, hospitals have limited resources and personnel to treat critical infections in intensive care units, which must be allocated effectively. To assist clinical staff in deciding which patients are in the greatest need of critical care, we develop a predictive model based on a publicly-available data set that is rich in clinical markers. We perform statistical analysis to determine which clinical markers strongly correlate with hospital admission, semi-intensive care, and intensive care for COVID-19 patients. We create a predictive model that will assist clinical personnel in determining COVID-19 patient prognosis. Additionally, we take a step towards a global framework for COVID-19 prognosis prediction by incorporating statistical data for geographically and ethnically diverse COVID--19 patient sets into our own model. To the best of our knowledge, this is the first model which does not exclusively utilize local data.
{"title":"Predicting Criticality in COVID-19 Patients","authors":"Roger A. Hallman, Anjali Chikkula, T. Prioleau","doi":"10.1145/3388440.3412463","DOIUrl":"https://doi.org/10.1145/3388440.3412463","url":null,"abstract":"The COVID-19 pandemic has infected millions of people around the world, spreading rapidly and causing a flood of patients that risk overwhelming clinical facilities. Whether in urban or rural areas, hospitals have limited resources and personnel to treat critical infections in intensive care units, which must be allocated effectively. To assist clinical staff in deciding which patients are in the greatest need of critical care, we develop a predictive model based on a publicly-available data set that is rich in clinical markers. We perform statistical analysis to determine which clinical markers strongly correlate with hospital admission, semi-intensive care, and intensive care for COVID-19 patients. We create a predictive model that will assist clinical personnel in determining COVID-19 patient prognosis. Additionally, we take a step towards a global framework for COVID-19 prognosis prediction by incorporating statistical data for geographically and ethnically diverse COVID--19 patient sets into our own model. To the best of our knowledge, this is the first model which does not exclusively utilize local data.","PeriodicalId":411338,"journal":{"name":"Proceedings of the 11th ACM International Conference on Bioinformatics, Computational Biology and Health Informatics","volume":"72 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-09-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126152895","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Jacob M. Schreiber, Timothy J. Durham, W. Noble, J. Bilmes
In the past decade, the use of high-throughput sequencing assays has allowed researchers to experimentally acquire thousands of functional measurements for each basepair in the human genome. Despite their value, these measurements are only a small fraction of the potential experiments that could be performed while also being too numerous to easily visualize or compute on. In a recent pair of publications [1,2], we address both of these challenges with a deep neural network tensor factorization method, Avocado, that compresses these measurements into dense, information-rich representations. We demonstrate that these learned representations can be used to impute, with high accuracy, the output of tens of thousands of functional experiments that have not yet been performed. Further, we show that, on a variety of genomics tasks, machine learning models that leverage these learned representations outperform those trained directly on the functional measurements. The code is publicly available at https://github.com/jmschrei/avocado.
{"title":"Avocado","authors":"Jacob M. Schreiber, Timothy J. Durham, W. Noble, J. Bilmes","doi":"10.1145/3388440.3414215","DOIUrl":"https://doi.org/10.1145/3388440.3414215","url":null,"abstract":"In the past decade, the use of high-throughput sequencing assays has allowed researchers to experimentally acquire thousands of functional measurements for each basepair in the human genome. Despite their value, these measurements are only a small fraction of the potential experiments that could be performed while also being too numerous to easily visualize or compute on. In a recent pair of publications [1,2], we address both of these challenges with a deep neural network tensor factorization method, Avocado, that compresses these measurements into dense, information-rich representations. We demonstrate that these learned representations can be used to impute, with high accuracy, the output of tens of thousands of functional experiments that have not yet been performed. Further, we show that, on a variety of genomics tasks, machine learning models that leverage these learned representations outperform those trained directly on the functional measurements. The code is publicly available at https://github.com/jmschrei/avocado.","PeriodicalId":411338,"journal":{"name":"Proceedings of the 11th ACM International Conference on Bioinformatics, Computational Biology and Health Informatics","volume":"75 1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-09-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"121274689","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Yuanqi Du, Anowarul Kabir, Liang Zhao, Amarda Shehu
Elucidating biologically-active protein structures remains a daunting task both in the wet and dry laboratory, and many proteins lack structural characterization. This lack of knowledge continues to motivate the development of computational methods for protein structure prediction. Methods are diverse in their approaches, and recent efforts have debuted deep learning-based methods for various sub-problems within the larger problem of protein structure prediction. In this paper, we focus on such a sub-problem, the reconstruction of three-dimensional structures consistent with given inter-atomic distances. Inspired by a recent architecture put forward in the larger context of generative frameworks, we design and evaluate a deep convolutional network model on experimentally- and computationally-obtained tertiary structures. Comparison with convex and stochastic optimization-based methods shows that the deep model is faster and similarly or more accurate, opening up several venues of further research to advance the larger problem of protein structure prediction.
{"title":"From Interatomic Distances to Protein Tertiary Structures with a Deep Convolutional Neural Network","authors":"Yuanqi Du, Anowarul Kabir, Liang Zhao, Amarda Shehu","doi":"10.1145/3388440.3414699","DOIUrl":"https://doi.org/10.1145/3388440.3414699","url":null,"abstract":"Elucidating biologically-active protein structures remains a daunting task both in the wet and dry laboratory, and many proteins lack structural characterization. This lack of knowledge continues to motivate the development of computational methods for protein structure prediction. Methods are diverse in their approaches, and recent efforts have debuted deep learning-based methods for various sub-problems within the larger problem of protein structure prediction. In this paper, we focus on such a sub-problem, the reconstruction of three-dimensional structures consistent with given inter-atomic distances. Inspired by a recent architecture put forward in the larger context of generative frameworks, we design and evaluate a deep convolutional network model on experimentally- and computationally-obtained tertiary structures. Comparison with convex and stochastic optimization-based methods shows that the deep model is faster and similarly or more accurate, opening up several venues of further research to advance the larger problem of protein structure prediction.","PeriodicalId":411338,"journal":{"name":"Proceedings of the 11th ACM International Conference on Bioinformatics, Computational Biology and Health Informatics","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-09-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114311887","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Saloni Agarwal, M. Abaker, Xinyi Zhang, O. Daescu, D. Barkauskas, E. Rudzinski, P. Leavey
A significant number of machine learning methods have been developed to identify major tumor types in histology images, yet much less is known about automatic classification of tumor subtypes. Rhabdomyosarcoma (RMS), the most common type of soft tissue cancer in children, has several subtypes, the most common being Embryonal, Alveolar, and Spindle Cell. Classifying RMS to the right subtype is critical, since subtypes are known to respond to different treatment protocols. Manual classification requires high expertise and is time consuming due to subtle variance in appearance of histopathology images. In this paper, we introduce and compare machine learning based architectures for automatic classification of Rhabdomyosarcoma into the three major subtypes, from whole slide images (WSI). For training purpose, we only know the class assigned to a WSI, having no manual annotations on the image, while most related work on tumor classification requires manual region or nuclei annotations on WSIs. To predict the class of a new WSI we first divide it into tiles, predict the class of each tile, then use thresholding with soft voting to convert tile level predictions to WSI level prediction. We obtain 94.87% WSI tumor subtype classification accuracy on a large and diverse test dataset. We achieve such accurate classification at 5X magnification level of WSIs, departing from related work, that uses 20X or 10X for best results. A direct advantage of our method is that both training and testing can be performed much faster computationally due to the lower image resolution.
{"title":"Rhabdomyosarcoma Histology Classification using Ensemble of Deep Learning Networks","authors":"Saloni Agarwal, M. Abaker, Xinyi Zhang, O. Daescu, D. Barkauskas, E. Rudzinski, P. Leavey","doi":"10.1145/3388440.3412486","DOIUrl":"https://doi.org/10.1145/3388440.3412486","url":null,"abstract":"A significant number of machine learning methods have been developed to identify major tumor types in histology images, yet much less is known about automatic classification of tumor subtypes. Rhabdomyosarcoma (RMS), the most common type of soft tissue cancer in children, has several subtypes, the most common being Embryonal, Alveolar, and Spindle Cell. Classifying RMS to the right subtype is critical, since subtypes are known to respond to different treatment protocols. Manual classification requires high expertise and is time consuming due to subtle variance in appearance of histopathology images. In this paper, we introduce and compare machine learning based architectures for automatic classification of Rhabdomyosarcoma into the three major subtypes, from whole slide images (WSI). For training purpose, we only know the class assigned to a WSI, having no manual annotations on the image, while most related work on tumor classification requires manual region or nuclei annotations on WSIs. To predict the class of a new WSI we first divide it into tiles, predict the class of each tile, then use thresholding with soft voting to convert tile level predictions to WSI level prediction. We obtain 94.87% WSI tumor subtype classification accuracy on a large and diverse test dataset. We achieve such accurate classification at 5X magnification level of WSIs, departing from related work, that uses 20X or 10X for best results. A direct advantage of our method is that both training and testing can be performed much faster computationally due to the lower image resolution.","PeriodicalId":411338,"journal":{"name":"Proceedings of the 11th ACM International Conference on Bioinformatics, Computational Biology and Health Informatics","volume":"43 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-09-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128159275","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
William Gasper, Kathryn M. Cooper, Nathan Cornelius, H. Ali, S. Bhowmick
With the increasing availability of various types of biological data and the ability to measure interrelationships among molecular elements, biological networks have quickly emerged as the go-to structure to model biological elements and relationships. However, there is not a large body of research that closely analyzes the properties of the various biological networks in ways that allow for the increased extraction of valuable information from these networks and establishes useful connections between network structures and corresponding biological properties. In particular, exploring the underlying graph properties of biological networks augments our understanding of biological organisms as complex systems. Understanding these properties is critical to the process of generating knowledge from biological network models. These properties become particularly interesting when they can be correlated with specific structural and functional qualities associated with the entities represented by the graph/network. Planarity may be especially important to understanding and identifying protein complexes, which are frequently subject to physical constraints that may prevent the constitutive protein components from interacting in such a way that the resulting graph abstraction is densely connected. In this work, we investigate the planarity of domain-domain interaction (DDI) graphs for S. cerevisiae protein complexes with validated three-dimensional structures. We found that the majority of these protein complexes were planar, even with the exclusion of complexes that had small DDI graphs with very few edges. We also found significant structural and functional differences between groups of complexes with planar and nonplanar DDI graphs. These results provide additional context for the study of protein complexes within the network model, and this additional context may be important for general knowledge generation, as well as for specific tasks like protein complex identification.
{"title":"Characterization of S. cerevisiae Protein Complexes by Representative DDI Graph Planarity","authors":"William Gasper, Kathryn M. Cooper, Nathan Cornelius, H. Ali, S. Bhowmick","doi":"10.1145/3388440.3412465","DOIUrl":"https://doi.org/10.1145/3388440.3412465","url":null,"abstract":"With the increasing availability of various types of biological data and the ability to measure interrelationships among molecular elements, biological networks have quickly emerged as the go-to structure to model biological elements and relationships. However, there is not a large body of research that closely analyzes the properties of the various biological networks in ways that allow for the increased extraction of valuable information from these networks and establishes useful connections between network structures and corresponding biological properties. In particular, exploring the underlying graph properties of biological networks augments our understanding of biological organisms as complex systems. Understanding these properties is critical to the process of generating knowledge from biological network models. These properties become particularly interesting when they can be correlated with specific structural and functional qualities associated with the entities represented by the graph/network. Planarity may be especially important to understanding and identifying protein complexes, which are frequently subject to physical constraints that may prevent the constitutive protein components from interacting in such a way that the resulting graph abstraction is densely connected. In this work, we investigate the planarity of domain-domain interaction (DDI) graphs for S. cerevisiae protein complexes with validated three-dimensional structures. We found that the majority of these protein complexes were planar, even with the exclusion of complexes that had small DDI graphs with very few edges. We also found significant structural and functional differences between groups of complexes with planar and nonplanar DDI graphs. These results provide additional context for the study of protein complexes within the network model, and this additional context may be important for general knowledge generation, as well as for specific tasks like protein complex identification.","PeriodicalId":411338,"journal":{"name":"Proceedings of the 11th ACM International Conference on Bioinformatics, Computational Biology and Health Informatics","volume":"6 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-09-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115741048","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Morgan Carothers, Joseph Gardi, Gianluca Gross, Tatsuki Kuze, Nuo Liu, Fiona Plunkett, Julia Qian, Yi-Chieh Wu
Given a gene tree, a species tree, and an association between their leaves, the maximum parsimony reconciliation (MPR) problem seeks to find a mapping of the gene tree to the species tree that explains their incongruity using a biological model of evolutionary events. Unfortunately, when simultaneously accounting for gene duplication, gene loss, and coalescence, the MPR problem is NP-hard. While an exact algorithm exists, it can be problematic to use in practice due to time and memory requirements. In this work, we present an integer linear programming (ILP) formulation for solving the MPR problem when considering duplications, losses, and coalescence. Our experimental results on a simulated data set of 12 Drosophila species shows that our new algorithm is both accurate and scalable. Furthermore, in contrast to the existing exact algorithm, our formulation allows users to limit the maximum runtime and thus trade-off accuracy and scalability, making it an attractive choice for phylogenetic pipelines.
{"title":"An Integer Linear Programming Solution for the Most Parsimonious Reconciliation Problem under the Duplication-Loss-Coalescence Model","authors":"Morgan Carothers, Joseph Gardi, Gianluca Gross, Tatsuki Kuze, Nuo Liu, Fiona Plunkett, Julia Qian, Yi-Chieh Wu","doi":"10.1145/3388440.3412474","DOIUrl":"https://doi.org/10.1145/3388440.3412474","url":null,"abstract":"Given a gene tree, a species tree, and an association between their leaves, the maximum parsimony reconciliation (MPR) problem seeks to find a mapping of the gene tree to the species tree that explains their incongruity using a biological model of evolutionary events. Unfortunately, when simultaneously accounting for gene duplication, gene loss, and coalescence, the MPR problem is NP-hard. While an exact algorithm exists, it can be problematic to use in practice due to time and memory requirements. In this work, we present an integer linear programming (ILP) formulation for solving the MPR problem when considering duplications, losses, and coalescence. Our experimental results on a simulated data set of 12 Drosophila species shows that our new algorithm is both accurate and scalable. Furthermore, in contrast to the existing exact algorithm, our formulation allows users to limit the maximum runtime and thus trade-off accuracy and scalability, making it an attractive choice for phylogenetic pipelines.","PeriodicalId":411338,"journal":{"name":"Proceedings of the 11th ACM International Conference on Bioinformatics, Computational Biology and Health Informatics","volume":"35 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-09-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128830194","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Advances in high-throughput microarray and RNA-sequencing technologies have lead to a rapid accumulation of gene expression data for various biological conditions across multiple species. Mining frequent gene modules from a set of multiple gene coexpression networks has applications in functional gene annotation and biomarker discovery. Biclustering algorithms have been proposed to allow for missing coexpression links. Existing approaches report a large number of edgesets which are computationally intensive to analyze, and have high overlap among the reported subnetworks. In this work, we propose an algorithm to mine frequent dense modules from multiple coexpression networks using an online data summarization method. Our algorithm mines a succinct set of representative subgraphs that have little overlap which reduces the downstream analysis of the reported modules. Experiments on human gene expression data show that the reported modules are biologically significant as evident by the high enrichment of GO molecular functions and KEGG pathways in the reported modules.
{"title":"Mining representative approximate frequent coexpression subnetworks","authors":"Sangmin Seo, Saeed Salem","doi":"10.1145/3388440.3415584","DOIUrl":"https://doi.org/10.1145/3388440.3415584","url":null,"abstract":"Advances in high-throughput microarray and RNA-sequencing technologies have lead to a rapid accumulation of gene expression data for various biological conditions across multiple species. Mining frequent gene modules from a set of multiple gene coexpression networks has applications in functional gene annotation and biomarker discovery. Biclustering algorithms have been proposed to allow for missing coexpression links. Existing approaches report a large number of edgesets which are computationally intensive to analyze, and have high overlap among the reported subnetworks. In this work, we propose an algorithm to mine frequent dense modules from multiple coexpression networks using an online data summarization method. Our algorithm mines a succinct set of representative subgraphs that have little overlap which reduces the downstream analysis of the reported modules. Experiments on human gene expression data show that the reported modules are biologically significant as evident by the high enrichment of GO molecular functions and KEGG pathways in the reported modules.","PeriodicalId":411338,"journal":{"name":"Proceedings of the 11th ACM International Conference on Bioinformatics, Computational Biology and Health Informatics","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-09-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131335514","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Altti Ilari Maarala, K. Pärn, J. Nuñez-Fontarnau, Keijo Heljanko
Massive whole-genome genotype reference panels now provide accurate and fast genotyping by imputation for high-resolution genome-wide association (GWA) studies. Imputation-assisted genotyping can increase the genomic coverage of genotypes and thus satisfy the resolution required in comprehensive GWA studies in a cost-effective manner. However, the imputation of missing genotypes from large reference panels is a compute-intensive process that requires high-performance computing (HPC). Although HPC uses extremely distributed and parallel computing, current imputation tools, and existing algorithms have not been developed to fully exploit the power of distributed computing. To this end, we have developed SparkBeagle, a scalable, fast, and accurate distributed genotype imputation tool based on popular Beagle software. SparkBeagle is designed for HPC and cloud computing environments and it is implemented on top of the Apache Spark distributed computing framework. We have carried out scalability experiments by imputing 64,976,316 variants of 2504 samples from the 1000 Genomes reference panel in the cloud. SparkBeagle shows near-linear scalability while increasing the number of computing nodes. A speedup of 30x was achieved with 40 nodes. The imputation time of the whole data set decreased from 565 minutes to 18 minutes compared to a single node parallel execution. Near identical imputation accuracy was measured in the concordance analysis between the original Beagle and the distributed SparkBeagle tool.
{"title":"SparkBeagle: Scalable Genotype Imputation from Distributed Whole-Genome Reference Panels in the Cloud","authors":"Altti Ilari Maarala, K. Pärn, J. Nuñez-Fontarnau, Keijo Heljanko","doi":"10.1145/3388440.3414860","DOIUrl":"https://doi.org/10.1145/3388440.3414860","url":null,"abstract":"Massive whole-genome genotype reference panels now provide accurate and fast genotyping by imputation for high-resolution genome-wide association (GWA) studies. Imputation-assisted genotyping can increase the genomic coverage of genotypes and thus satisfy the resolution required in comprehensive GWA studies in a cost-effective manner. However, the imputation of missing genotypes from large reference panels is a compute-intensive process that requires high-performance computing (HPC). Although HPC uses extremely distributed and parallel computing, current imputation tools, and existing algorithms have not been developed to fully exploit the power of distributed computing. To this end, we have developed SparkBeagle, a scalable, fast, and accurate distributed genotype imputation tool based on popular Beagle software. SparkBeagle is designed for HPC and cloud computing environments and it is implemented on top of the Apache Spark distributed computing framework. We have carried out scalability experiments by imputing 64,976,316 variants of 2504 samples from the 1000 Genomes reference panel in the cloud. SparkBeagle shows near-linear scalability while increasing the number of computing nodes. A speedup of 30x was achieved with 40 nodes. The imputation time of the whole data set decreased from 565 minutes to 18 minutes compared to a single node parallel execution. Near identical imputation accuracy was measured in the concordance analysis between the original Beagle and the distributed SparkBeagle tool.","PeriodicalId":411338,"journal":{"name":"Proceedings of the 11th ACM International Conference on Bioinformatics, Computational Biology and Health Informatics","volume":"7 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-09-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133811368","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Yue Li, Pratheeksha Nair, Zhi Wen, I. Chafi, A. Okhmatovskaia, G. Powell, Yannan Shen, D. Buckeridge
As the COVID-19 pandemic continues to unfold, understanding the global impact of non-pharmacological interventions (NPI) is important for formulating effective intervention strategies, particularly as many countries prepare for future waves. We used a machine learning approach to distill latent topics related to NPI from large-scale international news media. We hypothesize that these topics are informative about the timing and nature of implemented NPI, dependent on the source of the information (e.g., local news versus official government announcements) and the target countries. Given a set of latent topics associated with NPI (e.g., self-quarantine, social distancing, online education, etc), we assume that countries and media sources have different prior distributions over these topics, which are sampled to generate the news articles. To model the source-specific topic priors, we developed a semi-supervised, multi-source, dynamic, embedded topic model. Our model is able to simultaneously infer latent topics and learn a linear classifier to predict NPI labels using the topic mixtures as input for each news article. To learn these models, we developed an efficient end-to-end amortized variational inference algorithm. We applied our models to news data collected and labelled by the World Health Organization (WHO) and the Global Public Health Intelligence Network (GPHIN). Through comprehensive experiments, we observed superior topic quality and intervention prediction accuracy, compared to the baseline embedded topic models, which ignore information on media source and intervention labels. The inferred latent topics reveal distinct policies and media framing in different countries and media sources, and also characterize reaction to COVID-19 and NPI in a semantically meaningful manner. Our PyTorch code is available on Github (htps://github.com/li-lab-mcgill/covid19_media).
{"title":"Global Surveillance of COVID-19 by mining news media using a multi-source dynamic embedded topic model","authors":"Yue Li, Pratheeksha Nair, Zhi Wen, I. Chafi, A. Okhmatovskaia, G. Powell, Yannan Shen, D. Buckeridge","doi":"10.1145/3388440.3412418","DOIUrl":"https://doi.org/10.1145/3388440.3412418","url":null,"abstract":"As the COVID-19 pandemic continues to unfold, understanding the global impact of non-pharmacological interventions (NPI) is important for formulating effective intervention strategies, particularly as many countries prepare for future waves. We used a machine learning approach to distill latent topics related to NPI from large-scale international news media. We hypothesize that these topics are informative about the timing and nature of implemented NPI, dependent on the source of the information (e.g., local news versus official government announcements) and the target countries. Given a set of latent topics associated with NPI (e.g., self-quarantine, social distancing, online education, etc), we assume that countries and media sources have different prior distributions over these topics, which are sampled to generate the news articles. To model the source-specific topic priors, we developed a semi-supervised, multi-source, dynamic, embedded topic model. Our model is able to simultaneously infer latent topics and learn a linear classifier to predict NPI labels using the topic mixtures as input for each news article. To learn these models, we developed an efficient end-to-end amortized variational inference algorithm. We applied our models to news data collected and labelled by the World Health Organization (WHO) and the Global Public Health Intelligence Network (GPHIN). Through comprehensive experiments, we observed superior topic quality and intervention prediction accuracy, compared to the baseline embedded topic models, which ignore information on media source and intervention labels. The inferred latent topics reveal distinct policies and media framing in different countries and media sources, and also characterize reaction to COVID-19 and NPI in a semantically meaningful manner. Our PyTorch code is available on Github (htps://github.com/li-lab-mcgill/covid19_media).","PeriodicalId":411338,"journal":{"name":"Proceedings of the 11th ACM International Conference on Bioinformatics, Computational Biology and Health Informatics","volume":"25 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-09-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133721224","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","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：