Sing-Hoi Sze
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Sing-Hoi Sze has a total of 62 co-authors. Here are the top co-authors by number of shared publications:
- Aaron M. Tarone
Department of Entomology, Texas A&M University, College Station, TX 77843, USA
6 shared publications Gangman Yi
Department of Computer Science and Engineering, Gangneung-Wonju National University, Gangwon-do, South Korea
2 shared publications- Michael R. Thon
Department of Microbiology and Genetics, Institute for Agrobiotechnology Research (CIALE), University of Salamanca (USAL), Villamayor, Spain
2 shared publications - Albert Mulenga
Texas A & M University AgriLife Research, Department of Entomology, 2475 TAMU, College Station, TX 77843, USA
1 shared publication - Andrew J. Rosendale
University of Cincinnati
1 shared publication - Christopher J. Holmes
University of Cincinnati
1 shared publication
Latest Publications
- Open Access
- 9 September 2020
- Published by Springer Nature in BMC Genomics
- Vol. 21 (S9) , 1-10
Bayesian gamma-negative binomial modeling of single-cell RNA sequencing data
Background: Single-cell RNA sequencing (scRNA-seq) is a powerful profiling technique at the single-cell resolution. Appropriate analysis of scRNA-seq data can characterize molecular heterogeneity and shed light into the underlying cellular process to better understand development and disease mechanisms. The unique analytic challenge is to appropriately model highly over-dispersed scRNA-seq count data with prevalent dropouts (zero counts), making zero-inflated dimensionality reduction techniques popular for scRNA-seq data analyses. Employing zero-inflated distributions, however, may place extra emphasis on zero counts, leading to potential bias when identifying the latent structure of the data.Results: In this paper, we propose a fully generative hierarchical gamma-negative binomial (hGNB) model of scRNA-seq data, obviating the need for explicitly modeling zero inflation. At the same time, hGNB can naturally account for covariate effects at both the gene and cell levels to identify complex latent representations of scRNA-seq data, without the need for commonly adopted pre-processing steps such as normalization. Efficient Bayesian model inference is derived by exploiting conditional conjugacy via novel data augmentation techniques.Conclusion: Experimental results on both simulated data and several real-world scRNA-seq datasets suggest that hGNB is a powerful tool for cell cluster discovery as well as cell lineage inference.
- 9 September 2020
- Published by Springer Nature in BMC Genomics
- Vol. 21 (S9) , 1-10
Open AccessBayesian gamma-negative binomial modeling of single-cell RNA sequencing data
Background: Single-cell RNA sequencing (scRNA-seq) is a powerful profiling technique at the single-cell resolution. Appropriate analysis of scRNA-seq data can characterize molecular heterogeneity and shed light into the underlying cellular process to better understand development and disease mechanisms. The unique analytic challenge is to appropriately model highly over-dispersed scRNA-seq count data with prevalent dropouts (zero counts), making zero-inflated dimensionality reduction techniques popular for scRNA-seq data analyses. Employing zero-inflated distributions, however, may place extra emphasis on zero counts, leading to potential bias when identifying the latent structure of the data.Results: In this paper, we propose a fully generative hierarchical gamma-negative binomial (hGNB) model of scRNA-seq data, obviating the need for explicitly modeling zero inflation. At the same time, hGNB can naturally account for covariate effects at both the gene and cell levels to identify complex latent representations of scRNA-seq data, without the need for commonly adopted pre-processing steps such as normalization. Efficient Bayesian model inference is derived by exploiting conditional conjugacy via novel data augmentation techniques.Conclusion: Experimental results on both simulated data and several real-world scRNA-seq datasets suggest that hGNB is a powerful tool for cell cluster discovery as well as cell lineage inference.
- Open Access
- 6 June 2019
- Published by Springer Nature in BMC Genomics
- Vol. 20 (S5) , 1-14
Identifying similar transcripts in a related organism from de Bruijn graphs of RNA-Seq data, with applications to the study of salt and waterlogging tolerance in Melilotus
A popular strategy to study alternative splicing in non-model organisms starts from sequencing the entire transcriptome, then assembling the reads by using de novo transcriptome assembly algorithms to obtain predicted transcripts. A similarity search algorithm is then applied to a related organism to infer possible function of these predicted transcripts. While some of these predictions may be inaccurate and transcripts with low coverage are often missed, we observe that it is possible to obtain a more complete set of transcripts to facilitate possible functional assignments by starting the search from the intermediate de Bruijn graph that contains all branching possibilities. We develop an algorithm to extract similar transcripts in a related organism by starting the search from the de Bruijn graph that represents the transcriptome instead of from predicted transcripts. We show that our algorithm is able to recover more similar transcripts than existing algorithms, with large improvements in obtaining longer transcripts and a finer resolution of isoforms. We apply our algorithm to study salt and waterlogging tolerance in two Melilotus species by constructing new RNA-Seq libraries. We have developed an algorithm to identify paths in the de Bruijn graph that correspond to similar transcripts in a related organism directly. Our strategy bypasses the transcript prediction step in RNA-Seq data and makes use of support from evolutionary information.
- 6 June 2019
- Published by Springer Nature in BMC Genomics
- Vol. 20 (S5) , 1-14
Open AccessIdentifying similar transcripts in a related organism from de Bruijn graphs of RNA-Seq data, with applications to the study of salt and waterlogging tolerance in Melilotus
A popular strategy to study alternative splicing in non-model organisms starts from sequencing the entire transcriptome, then assembling the reads by using de novo transcriptome assembly algorithms to obtain predicted transcripts. A similarity search algorithm is then applied to a related organism to infer possible function of these predicted transcripts. While some of these predictions may be inaccurate and transcripts with low coverage are often missed, we observe that it is possible to obtain a more complete set of transcripts to facilitate possible functional assignments by starting the search from the intermediate de Bruijn graph that contains all branching possibilities. We develop an algorithm to extract similar transcripts in a related organism by starting the search from the de Bruijn graph that represents the transcriptome instead of from predicted transcripts. We show that our algorithm is able to recover more similar transcripts than existing algorithms, with large improvements in obtaining longer transcripts and a finer resolution of isoforms. We apply our algorithm to study salt and waterlogging tolerance in two Melilotus species by constructing new RNA-Seq libraries. We have developed an algorithm to identify paths in the de Bruijn graph that correspond to similar transcripts in a related organism directly. Our strategy bypasses the transcript prediction step in RNA-Seq data and makes use of support from evolutionary information.
- Open Access
Functional genomics of the stable fly, Stomoxys calcitrans , reveals mechanisms underlying reproduction, host interactions, and novel targets for pest control
Background: The stable fly, Stomoxys calcitrans , is a major blood-feeding pest of livestock that has near worldwide distribution, causing an annual cost of over $2 billion for control and product loss in the United States alone. Control of these flies has been limited to increased sanitary management practices and insecticide application for suppressing larval stages. Few genetic and molecular resources are available to help in developing novel methods for controlling stable flies. Results: This study examines stable fly biology by utilizing a combination of high-quality genome sequencing, microbiome analyses, and RNA-seq analyses targeting multiple developmental stages and tissues. In conjunction, manual curation of over 1600 genes was used to examine gene content related to stable fly reproduction, interactions with their host, host-microbe dynamics, and putative routes for control. Most notable was establishment of reproduction-associated genes and identification of expanded vision, chemosensation, immune repertoire, and metabolic detoxification pathway gene families. Conclusions: The combined sequencing, assembly, and curation of the male stable fly genome followed by RNA-seq and downstream analyses provide insights necessary to understand the biology of this important pest. These resources and knowledge will provide the groundwork for expanding the tools available to control stable fly infestations. The close relationship of Stomoxys to other blood-feeding ( Glossina ) and non-blood-feeding flies (medflies, Drosophila , house flies) will allow for understanding the evolution of blood feeding among Cyclorrhapha flies.
Open AccessFunctional genomics of the stable fly, Stomoxys calcitrans , reveals mechanisms underlying reproduction, host interactions, and novel targets for pest control
Background: The stable fly, Stomoxys calcitrans , is a major blood-feeding pest of livestock that has near worldwide distribution, causing an annual cost of over $2 billion for control and product loss in the United States alone. Control of these flies has been limited to increased sanitary management practices and insecticide application for suppressing larval stages. Few genetic and molecular resources are available to help in developing novel methods for controlling stable flies. Results: This study examines stable fly biology by utilizing a combination of high-quality genome sequencing, microbiome analyses, and RNA-seq analyses targeting multiple developmental stages and tissues. In conjunction, manual curation of over 1600 genes was used to examine gene content related to stable fly reproduction, interactions with their host, host-microbe dynamics, and putative routes for control. Most notable was establishment of reproduction-associated genes and identification of expanded vision, chemosensation, immune repertoire, and metabolic detoxification pathway gene families. Conclusions: The combined sequencing, assembly, and curation of the male stable fly genome followed by RNA-seq and downstream analyses provide insights necessary to understand the biology of this important pest. These resources and knowledge will provide the groundwork for expanding the tools available to control stable fly infestations. The close relationship of Stomoxys to other blood-feeding ( Glossina ) and non-blood-feeding flies (medflies, Drosophila , house flies) will allow for understanding the evolution of blood feeding among Cyclorrhapha flies.
- Open Access
- 6 December 2017
- Published by Springer Nature in BMC Genomics
- Vol. 18 (S10) , 895
A divide-and-conquer algorithm for large-scale de novo transcriptome assembly through combining small assemblies from existing algorithms
While the continued development of high-throughput sequencing has facilitated studies of entire transcriptomes in non-model organisms, the incorporation of an increasing amount of RNA-Seq libraries has made de novo transcriptome assembly difficult. Although algorithms that can assemble a large amount of RNA-Seq data are available, they are generally very memory-intensive and can only be used to construct small assemblies. We develop a divide-and-conquer strategy that allows these algorithms to be utilized, by subdividing a large RNA-Seq data set into small libraries. Each individual library is assembled independently by an existing algorithm, and a merging algorithm is developed to combine these assemblies by picking a subset of high quality transcripts to form a large transcriptome. When compared to existing algorithms that return a single assembly directly, this strategy achieves comparable or increased accuracy as memory-efficient algorithms that can be used to process a large amount of RNA-Seq data, and comparable or decreased accuracy as memory-intensive algorithms that can only be used to construct small assemblies. Our divide-and-conquer strategy allows memory-intensive de novo transcriptome assembly algorithms to be utilized to construct large assemblies.
- 6 December 2017
- Published by Springer Nature in BMC Genomics
- Vol. 18 (S10) , 895
Open AccessA divide-and-conquer algorithm for large-scale de novo transcriptome assembly through combining small assemblies from existing algorithms
While the continued development of high-throughput sequencing has facilitated studies of entire transcriptomes in non-model organisms, the incorporation of an increasing amount of RNA-Seq libraries has made de novo transcriptome assembly difficult. Although algorithms that can assemble a large amount of RNA-Seq data are available, they are generally very memory-intensive and can only be used to construct small assemblies. We develop a divide-and-conquer strategy that allows these algorithms to be utilized, by subdividing a large RNA-Seq data set into small libraries. Each individual library is assembled independently by an existing algorithm, and a merging algorithm is developed to combine these assemblies by picking a subset of high quality transcripts to form a large transcriptome. When compared to existing algorithms that return a single assembly directly, this strategy achieves comparable or increased accuracy as memory-efficient algorithms that can be used to process a large amount of RNA-Seq data, and comparable or decreased accuracy as memory-intensive algorithms that can only be used to construct small assemblies. Our divide-and-conquer strategy allows memory-intensive de novo transcriptome assembly algorithms to be utilized to construct large assemblies.
- Open Access
- 24 May 2017
- Published by Springer Nature in BMC Genomics
- Vol. 18 (4) , 1-11
A scalable and memory-efficient algorithm for de novo transcriptome assembly of non-model organisms
With increased availability of de novo assembly algorithms, it is feasible to study entire transcriptomes of non-model organisms. While algorithms are available that are specifically designed for performing transcriptome assembly from high-throughput sequencing data, they are very memory-intensive, limiting their applications to small data sets with few libraries. We develop a transcriptome assembly algorithm that recovers alternatively spliced isoforms and expression levels while utilizing as many RNA-Seq libraries as possible that contain hundreds of gigabases of data. New techniques are developed so that computations can be performed on a computing cluster with moderate amount of physical memory. Our strategy minimizes memory consumption while simultaneously obtaining comparable or improved accuracy over existing algorithms. It provides support for incremental updates of assemblies when new libraries become available.
- 24 May 2017
- Published by Springer Nature in BMC Genomics
- Vol. 18 (4) , 1-11
Open AccessA scalable and memory-efficient algorithm for de novo transcriptome assembly of non-model organisms
With increased availability of de novo assembly algorithms, it is feasible to study entire transcriptomes of non-model organisms. While algorithms are available that are specifically designed for performing transcriptome assembly from high-throughput sequencing data, they are very memory-intensive, limiting their applications to small data sets with few libraries. We develop a transcriptome assembly algorithm that recovers alternatively spliced isoforms and expression levels while utilizing as many RNA-Seq libraries as possible that contain hundreds of gigabases of data. New techniques are developed so that computations can be performed on a computing cluster with moderate amount of physical memory. Our strategy minimizes memory consumption while simultaneously obtaining comparable or improved accuracy over existing algorithms. It provides support for incremental updates of assemblies when new libraries become available.
- Open Access
- 10 November 2015
- Published by Springer Nature in BMC Genomics
- Vol. 16 (S11) , S5
Heuristic pairwise alignment of de Bruijn graphs to facilitate simultaneous transcript discovery in related organisms from RNA-Seq data
The advance of high-throughput sequencing has made it possible to obtain new transcriptomes and study splicing mechanisms in non-model organisms. In these studies, there is often a need to investigate the transcriptomes of two related organisms at the same time in order to find the similarities and differences between them. The traditional approach to address this problem is to perform de novo transcriptome assemblies to obtain predicted transcripts for these organisms independently and then employ similarity comparison algorithms to study them.
- 10 November 2015
- Published by Springer Nature in BMC Genomics
- Vol. 16 (S11) , S5
Open AccessHeuristic pairwise alignment of de Bruijn graphs to facilitate simultaneous transcript discovery in related organisms from RNA-Seq data
The advance of high-throughput sequencing has made it possible to obtain new transcriptomes and study splicing mechanisms in non-model organisms. In these studies, there is often a need to investigate the transcriptomes of two related organisms at the same time in order to find the similarities and differences between them. The traditional approach to address this problem is to perform de novo transcriptome assemblies to obtain predicted transcripts for these organisms independently and then employ similarity comparison algorithms to study them.
- Open Access
- 14 July 2014
- Published by Springer Nature in BMC Genomics
- Vol. 15 (S5) , S6
A memory-efficient algorithm to obtain splicing graphs and de novoexpression estimates from de Bruijn graphs of RNA-Seq data
The recent advance of high-throughput sequencing makes it feasible to study entire transcriptomes through the application of de novo sequence assembly algorithms. While a popular strategy is to first construct an intermediate de Bruijn graph structure to represent the transcriptome, an additional step is needed to construct predicted transcripts from the graph.
- 14 July 2014
- Published by Springer Nature in BMC Genomics
- Vol. 15 (S5) , S6
Open AccessA memory-efficient algorithm to obtain splicing graphs and de novoexpression estimates from de Bruijn graphs of RNA-Seq data
The recent advance of high-throughput sequencing makes it feasible to study entire transcriptomes through the application of de novo sequence assembly algorithms. While a popular strategy is to first construct an intermediate de Bruijn graph structure to represent the transcriptome, an additional step is needed to construct predicted transcripts from the graph.
- Open Access
- 24 June 2014
- Published by Springer Nature in BMC Genomics
- Vol. 15 (1) , 1-30
A 24-48 h fed Amblyomma americanum tick saliva immuno-proteome
Multiple tick saliva proteins, the majority of which are unknown, confer tick resistance in repeatedly infested animals. The objective of this study was to identify the 24-48 h fed Amblyomma americanum tick saliva immuno-proteome. The 24-48 h tick-feeding phase is critical to tick parasitism as it precedes important events in tick biology, blood meal feeding and disease agent transmission. Fed male, 24 and 96 h fed female phage display cDNA expression libraries were biopanned using rabbit antibodies to 24 and 48 h fed A. americanum female tick saliva proteins. Biopanned immuno-cDNA libraries were subjected to next generation sequencing, de novo assembly, and bioinformatic analysis. More than 800 transcripts that code for 24-48 h fed A. americanum immuno-proteins are described. Of the 895 immuno-proteins, 52% (464/895) were provisionally identified based on matches in GenBank. Of these, ~19% (86/464) show high level of identity to other tick hypothetical proteins, and the rest include putative proteases (serine, cysteine, leukotriene A-4 hydrolase, carboxypeptidases, and metalloproteases), protease inhibitors (serine and cysteine protease inhibitors, tick carboxypeptidase inhibitor), and transporters and/or ligand binding proteins (histamine binding/lipocalin, fatty acid binding, calreticulin, hemelipoprotein, IgG binding protein, ferritin, insulin-like growth factor binding proteins, and evasin). Others include enzymes (glutathione transferase, cytochrome oxidase, protein disulfide isomerase), ribosomal proteins, and those of miscellaneous functions (histamine release factor, selenoproteins, tetraspanin, defensin, heat shock proteins). Data here demonstrate that A. americanum secretes a complex cocktail of immunogenic tick saliva proteins during the first 24-48 h of feeding. Of significance, previously validated immunogenic tick saliva proteins including AV422 protein, calreticulin, histamine release factor, histamine binding/lipocalins, selenoproteins, and paramyosin were identified in this screen, supporting the specificity of the approach in this study. While descriptive, this study opens opportunities for in-depth tick feeding physiology studies.
- 24 June 2014
- Published by Springer Nature in BMC Genomics
- Vol. 15 (1) , 1-30
Open AccessA 24-48 h fed Amblyomma americanum tick saliva immuno-proteome
Multiple tick saliva proteins, the majority of which are unknown, confer tick resistance in repeatedly infested animals. The objective of this study was to identify the 24-48 h fed Amblyomma americanum tick saliva immuno-proteome. The 24-48 h tick-feeding phase is critical to tick parasitism as it precedes important events in tick biology, blood meal feeding and disease agent transmission. Fed male, 24 and 96 h fed female phage display cDNA expression libraries were biopanned using rabbit antibodies to 24 and 48 h fed A. americanum female tick saliva proteins. Biopanned immuno-cDNA libraries were subjected to next generation sequencing, de novo assembly, and bioinformatic analysis. More than 800 transcripts that code for 24-48 h fed A. americanum immuno-proteins are described. Of the 895 immuno-proteins, 52% (464/895) were provisionally identified based on matches in GenBank. Of these, ~19% (86/464) show high level of identity to other tick hypothetical proteins, and the rest include putative proteases (serine, cysteine, leukotriene A-4 hydrolase, carboxypeptidases, and metalloproteases), protease inhibitors (serine and cysteine protease inhibitors, tick carboxypeptidase inhibitor), and transporters and/or ligand binding proteins (histamine binding/lipocalin, fatty acid binding, calreticulin, hemelipoprotein, IgG binding protein, ferritin, insulin-like growth factor binding proteins, and evasin). Others include enzymes (glutathione transferase, cytochrome oxidase, protein disulfide isomerase), ribosomal proteins, and those of miscellaneous functions (histamine release factor, selenoproteins, tetraspanin, defensin, heat shock proteins). Data here demonstrate that A. americanum secretes a complex cocktail of immunogenic tick saliva proteins during the first 24-48 h of feeding. Of significance, previously validated immunogenic tick saliva proteins including AV422 protein, calreticulin, histamine release factor, histamine binding/lipocalins, selenoproteins, and paramyosin were identified in this screen, supporting the specificity of the approach in this study. While descriptive, this study opens opportunities for in-depth tick feeding physiology studies.
- 1 December 2012
- Published by SAGE Publications in Journal of Computational Biology
- Vol. 19 (12) , 1284-1294
Identifying Complexes from Protein Interaction Networks According to Different Types of Neighborhood Density
To facilitate the realization of biological functions, proteins are often organized into complexes. While computational techniques are used to predict these complexes, detailed understanding of their organization remains inadequate. Apart from complexes that reside in very dense regions of a protein interaction network in which most algorithms are able to identify, we observe that many other complexes, while not residing in very dense regions, reside in regions with low neighborhood density. We develop an algorithm for identifying protein complexes by considering these two types of complexes separately. We test our algorithm on a few yeast protein interaction networks, and show that our algorithm is able to identify complexes more accurately than existing algorithms. A software program NDComplex for implementing the algorithm is available at http://faculty.cse.tamu.edu/shsze/ndcomplex.
- 1 December 2012
- Published by SAGE Publications in Journal of Computational Biology
- Vol. 19 (12) , 1284-1294
Identifying Complexes from Protein Interaction Networks According to Different Types of Neighborhood Density
To facilitate the realization of biological functions, proteins are often organized into complexes. While computational techniques are used to predict these complexes, detailed understanding of their organization remains inadequate. Apart from complexes that reside in very dense regions of a protein interaction network in which most algorithms are able to identify, we observe that many other complexes, while not residing in very dense regions, reside in regions with low neighborhood density. We develop an algorithm for identifying protein complexes by considering these two types of complexes separately. We test our algorithm on a few yeast protein interaction networks, and show that our algorithm is able to identify complexes more accurately than existing algorithms. A software program NDComplex for implementing the algorithm is available at http://faculty.cse.tamu.edu/shsze/ndcomplex.
- 1 August 2012
- Published by SAGE Publications in Journal of Computational Biology
- Vol. 19 (8) , 957-967
Supervised Protein Family Classification and New Family Construction
The goal of protein family classification is to group proteins into families so that proteins within the same family have common function or are related by ancestry. While supervised classification algorithms are available for this purpose, most of these approaches focus on assigning unclassified proteins to known families but do not allow for progressive construction of new families from proteins that cannot be assigned. Although unsupervised clustering algorithms are also available, they do not make use of information from known families. By computing similarities between proteins based on pairwise sequence comparisons, we develop supervised classification algorithms that achieve improved accuracy over previous approaches while allowing for construction of new families. We show that our algorithm has higher accuracy rate and lower mis-classification rate when compared to algorithms that are based on the use of multiple sequence alignments and hidden Markov models, and our algorithm performs well even on families with very few proteins and on families with low sequence similarity. A software program implementing the algorithm (SClassify) is available online ( http://faculty.cse.tamu.edu/shsze/sclassify ).
- 1 August 2012
- Published by SAGE Publications in Journal of Computational Biology
- Vol. 19 (8) , 957-967
Supervised Protein Family Classification and New Family Construction
The goal of protein family classification is to group proteins into families so that proteins within the same family have common function or are related by ancestry. While supervised classification algorithms are available for this purpose, most of these approaches focus on assigning unclassified proteins to known families but do not allow for progressive construction of new families from proteins that cannot be assigned. Although unsupervised clustering algorithms are also available, they do not make use of information from known families. By computing similarities between proteins based on pairwise sequence comparisons, we develop supervised classification algorithms that achieve improved accuracy over previous approaches while allowing for construction of new families. We show that our algorithm has higher accuracy rate and lower mis-classification rate when compared to algorithms that are based on the use of multiple sequence alignments and hidden Markov models, and our algorithm performs well even on families with very few proteins and on families with low sequence similarity. A software program implementing the algorithm (SClassify) is available online ( http://faculty.cse.tamu.edu/shsze/sclassify ).