Seyoon Ko
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Seyoon Ko has a total of 26 co-authors. Here are the top co-authors by number of shared publications:
- Hua Zhou
Department of Biostatistics, University of California, Los Angeles, Los Angeles, USA
6 shared publications - Eric M. Sobel
Department of Human Genetics, David Geffen School of Medicine, University of California, Los Angeles, Los Angeles, California
4 shared publications - Kenneth Lange
Department of Computational Medicine, University of California, Los Angeles, Los Angeles, USA
4 shared publications - Benjamin B. Chu
Affiliation Department of Biomedical Data Science, Stanford University, Stanford, California, United States of America
3 shared publications Chidera Okenwa
Department of Mathematics, University of California, Berkeley, Berkeley, CA 94720, USA
2 shared publicationsDaniel Peterson
Department of Mathematics, Brigham Young University, Provo, UT 84602, USA
2 shared publications
Latest Publications
- Open Access
- 10 December 2024
- Published by American Association for the Advancement of Science (AAAS) in Computational and Structural Biotechnology Journal
- Vol. 23, 4384-4395
Estimation of genetic admixture proportions via haplotypes
Estimation of ancestral admixture is essential for creating personal genealogies, studying human history, and conducting genome-wide association studies (GWAS). The following three primary methods exist for estimating admixture coefficients. The frequentist approach directly maximizes the binomial loglikelihood. The Bayesian approach adds a reasonable prior and samples the posterior distribution. Finally, the nonparametric approach decomposes the genotype matrix algebraically. Each approach scales successfully to datasets with a million individuals and a million single nucleotide polymorphisms (SNPs). Despite their variety, all current approaches assume independence between SNPs. To achieve independence requires performing LD (linkage disequilibrium) filtering before analysis. Unfortunately, this tactic loses valuable information and usually retains many SNPs still in LD. The present paper explores the option of explicitly incorporating haplotypes in ancestry estimation. Our program, HaploADMIXTURE, operates on adjacent SNP pairs and jointly estimates their haplotype frequencies along with admixture coefficients. This more complex strategy takes advantage of the rich information available in haplotypes and ultimately yields better admixture estimates and better clustering of real populations in curated datasets.
- 10 December 2024
- Published by American Association for the Advancement of Science (AAAS) in Computational and Structural Biotechnology Journal
- Vol. 23, 4384-4395
Open AccessEstimation of genetic admixture proportions via haplotypes
Estimation of ancestral admixture is essential for creating personal genealogies, studying human history, and conducting genome-wide association studies (GWAS). The following three primary methods exist for estimating admixture coefficients. The frequentist approach directly maximizes the binomial loglikelihood. The Bayesian approach adds a reasonable prior and samples the posterior distribution. Finally, the nonparametric approach decomposes the genotype matrix algebraically. Each approach scales successfully to datasets with a million individuals and a million single nucleotide polymorphisms (SNPs). Despite their variety, all current approaches assume independence between SNPs. To achieve independence requires performing LD (linkage disequilibrium) filtering before analysis. Unfortunately, this tactic loses valuable information and usually retains many SNPs still in LD. The present paper explores the option of explicitly incorporating haplotypes in ancestry estimation. Our program, HaploADMIXTURE, operates on adjacent SNP pairs and jointly estimates their haplotype frequencies along with admixture coefficients. This more complex strategy takes advantage of the rich information available in haplotypes and ultimately yields better admixture estimates and better clustering of real populations in curated datasets.
- Open Access
Estimation of Genetic Admixture Proportions via Haplotypes
Estimation of ethnic admixture is essential for creating personal genealogies, studying human history, and conducting genome-wide association studies (GWAS). Three methods exist for estimating admixture coefficients. The frequentist approach directly maximizes the binomial loglikelihood. The Bayesian approach adds a reasonable prior and samples the posterior distribution. Finally, the nonparametric approach decomposes the genotype matrix algebraically. Each approach scales successfully to data sets with a million individuals and a million single nucleotide polymorphisms (SNPs). Despite their variety, all current approaches assume independence between SNPs. To achieve independence requires performing LD (linkage disequilibrium) filtering before analysis. Unfortunately, this tactic loses valuable information and usually retains many SNPs still in LD. The present paper explores the option of explicitly incorporating haplotypes in ancestry estimation. Our program, HaploADMIXTURE, operates on adjacent SNP pairs and jointly estimates their haplotype frequencies along with admixture coefficients. This more complex strategy takes advantage of the rich information available in haplotypes and ultimately yields better admixture estimates and better clustering of real populations in curated data sets.
Open AccessEstimation of Genetic Admixture Proportions via Haplotypes
Estimation of ethnic admixture is essential for creating personal genealogies, studying human history, and conducting genome-wide association studies (GWAS). Three methods exist for estimating admixture coefficients. The frequentist approach directly maximizes the binomial loglikelihood. The Bayesian approach adds a reasonable prior and samples the posterior distribution. Finally, the nonparametric approach decomposes the genotype matrix algebraically. Each approach scales successfully to data sets with a million individuals and a million single nucleotide polymorphisms (SNPs). Despite their variety, all current approaches assume independence between SNPs. To achieve independence requires performing LD (linkage disequilibrium) filtering before analysis. Unfortunately, this tactic loses valuable information and usually retains many SNPs still in LD. The present paper explores the option of explicitly incorporating haplotypes in ancestry estimation. Our program, HaploADMIXTURE, operates on adjacent SNP pairs and jointly estimates their haplotype frequencies along with admixture coefficients. This more complex strategy takes advantage of the rich information available in haplotypes and ultimately yields better admixture estimates and better clustering of real populations in curated data sets.
- Open Access
- 6 January 2023
- Published by Elsevier in The American Journal of Human Genetics
- Vol. 110 (2) , 314-325
Unsupervised discovery of ancestry-informative markers and genetic admixture proportions in biobank-scale datasets
- 6 January 2023
- Published by Elsevier in The American Journal of Human Genetics
- Vol. 110 (2) , 314-325
Open AccessUnsupervised discovery of ancestry-informative markers and genetic admixture proportions in biobank-scale datasets
- Open Access
Unsupervised Discovery of Ancestry Informative Markers and Genetic Admixture Proportions in Biobank-Scale Data Sets
Admixture estimation plays a crucial role in ancestry inference and genomewide association studies (GWAS). Computer programs such as ADMIXTURE and STRUCTURE are commonly employed to estimate the admixture proportions of sample individuals. However, these programs can be overwhelmed by the computational burdens imposed by the 10 5 to 10 6 samples and millions of markers commonly found in modern biobanks. An attractive strategy is to run these programs on a set of ancestry informative SNP markers (AIMs) that exhibit substantially different frequencies across populations. Unfortunately, existing methods for identifying AIMs require knowing ancestry labels for a subset of the sample. This supervised learning approach creates a chicken and the egg scenario. In this paper, we present an unsupervised, scalable framework that seamlessly carries out AIM selection and likelihood-based estimation of admixture proportions. Our simulated and real data examples show that this approach is scalable to modern biobank data sets. Our implementation of the method is called OpenADMIXTURE.
Open AccessUnsupervised Discovery of Ancestry Informative Markers and Genetic Admixture Proportions in Biobank-Scale Data Sets
Admixture estimation plays a crucial role in ancestry inference and genomewide association studies (GWAS). Computer programs such as ADMIXTURE and STRUCTURE are commonly employed to estimate the admixture proportions of sample individuals. However, these programs can be overwhelmed by the computational burdens imposed by the 10 5 to 10 6 samples and millions of markers commonly found in modern biobanks. An attractive strategy is to run these programs on a set of ancestry informative SNP markers (AIMs) that exhibit substantially different frequencies across populations. Unfortunately, existing methods for identifying AIMs require knowing ancestry labels for a subset of the sample. This supervised learning approach creates a chicken and the egg scenario. In this paper, we present an unsupervised, scalable framework that seamlessly carries out AIM selection and likelihood-based estimation of admixture proportions. Our simulated and real data examples show that this approach is scalable to modern biobank data sets. Our implementation of the method is called OpenADMIXTURE.
- 21 December 2021
- Published by Institute of Mathematical Statistics in The Annals of Applied Statistics
- Vol. 15 (4) , 1652-1672
VCSEL: Prioritizing SNP-set by penalized variance component selection
Single nucleotide polymorphism (SNP) set analysis aggregates both common and rare variants and tests for association between phenotype(s) of interest and a set. However, multiple SNP-sets, such as genes, pathways, or sliding windows are usually investigated across the whole genome in which all groups are tested separately, followed by multiple testing adjustments. We propose a novel method to prioritize SNP-sets in a joint multivariate variance component model. Each SNP-set corresponds to a variance component (or kernel), and model selection is achieved by incorporating either convex or nonconvex penalties. The uniqueness of this variance component selection framework, which we call VCSEL, is that it naturally encompasses multivariate traits (VCSEL-M) and SNP-set-treatment or -environment interactions (VCSEL-I). We devise an optimization algorithm scalable to many variance components, based on the majorization-minimization (MM) principle. Simulation studies demonstrate the superiority of our methods in model selection performance, as measured by the area under the precision-recall (PR) curve, compared to the commonly used marginal testing and group penalization methods. Finally, we apply our methods to a real pharmacogenomics study and a real whole exome sequencing study. Some top ranked genes by VCSEL are detected as insignificant by the marginal test methods which emphasizes formal inference of individual genes with a strict significance threshold. This provides alternative insights for biologists to prioritize follow-up studies and develop polygenic risk score models.
- 21 December 2021
- Published by Institute of Mathematical Statistics in The Annals of Applied Statistics
- Vol. 15 (4) , 1652-1672
VCSEL: Prioritizing SNP-set by penalized variance component selection
Single nucleotide polymorphism (SNP) set analysis aggregates both common and rare variants and tests for association between phenotype(s) of interest and a set. However, multiple SNP-sets, such as genes, pathways, or sliding windows are usually investigated across the whole genome in which all groups are tested separately, followed by multiple testing adjustments. We propose a novel method to prioritize SNP-sets in a joint multivariate variance component model. Each SNP-set corresponds to a variance component (or kernel), and model selection is achieved by incorporating either convex or nonconvex penalties. The uniqueness of this variance component selection framework, which we call VCSEL, is that it naturally encompasses multivariate traits (VCSEL-M) and SNP-set-treatment or -environment interactions (VCSEL-I). We devise an optimization algorithm scalable to many variance components, based on the majorization-minimization (MM) principle. Simulation studies demonstrate the superiority of our methods in model selection performance, as measured by the area under the precision-recall (PR) curve, compared to the commonly used marginal testing and group penalization methods. Finally, we apply our methods to a real pharmacogenomics study and a real whole exome sequencing study. Some top ranked genes by VCSEL are detected as insignificant by the marginal test methods which emphasizes formal inference of individual genes with a strict significance threshold. This provides alternative insights for biologists to prioritize follow-up studies and develop polygenic risk score models.
- Open Access
Multivariate Genomewide Association Analysis by Iterative Hard Thresholding
1 In genome-wide association studies (GWAS), analyzing multiple correlated traits is potentially superior to conducting multiple univariate analyses. Standard methods for multivariate GWAS operate marker-by-marker and are computationally intensive. We present a penalized regression algorithm for multivariate GWAS based on iterative hard thresholding (IHT) and implement it in a convenient Julia package MendelIHT.jl ( https://github.com/OpenMendel/MendelIHT.jl ). In simulation studies with up to 100 traits, IHT exhibits similar true positive rates, smaller false positive rates, and faster execution times than GEMMA ’s linear mixed models and mv-PLINK ’s canonical correlation analysis. On UK Biobank data, our IHT software completed a 3-trait joint analysis in 20 hours and an 18-trait joint analysis in 53 hours, requiring up to 80GB of computer memory. In short, our software enables geneticists to fit a single regression model that simultaneously considers the effect of all SNPs and dozens of traits.
Open AccessMultivariate Genomewide Association Analysis by Iterative Hard Thresholding
1 In genome-wide association studies (GWAS), analyzing multiple correlated traits is potentially superior to conducting multiple univariate analyses. Standard methods for multivariate GWAS operate marker-by-marker and are computationally intensive. We present a penalized regression algorithm for multivariate GWAS based on iterative hard thresholding (IHT) and implement it in a convenient Julia package MendelIHT.jl ( https://github.com/OpenMendel/MendelIHT.jl ). In simulation studies with up to 100 traits, IHT exhibits similar true positive rates, smaller false positive rates, and faster execution times than GEMMA ’s linear mixed models and mv-PLINK ’s canonical correlation analysis. On UK Biobank data, our IHT software completed a 3-trait joint analysis in 20 hours and an 18-trait joint analysis in 53 hours, requiring up to 80GB of computer memory. In short, our software enables geneticists to fit a single regression model that simultaneously considers the effect of all SNPs and dozens of traits.
- Open Access
- 21 July 2021
- Published by Oxford University Press (OUP) in Bioinformatics
- Vol. 37 (24) , 4756-4763
A fast data-driven method for genotype imputation, phasing and local ancestry inference: MendelImpute.jl
Current methods for genotype imputation and phasing exploit the volume of data in haplotype reference panels and rely on hidden Markov models (HMMs). Existing programs all have essentially the same imputation accuracy, are computationally intensive and generally require prephasing the typed markers. We introduce a novel data-mining method for genotype imputation and phasing that substitutes highly efficient linear algebra routines for HMM calculations. This strategy, embodied in our Julia program MendelImpute.jl, avoids explicit assumptions about recombination and population structure while delivering similar prediction accuracy, better memory usage and an order of magnitude or better run-times compared to the fastest competing method. MendelImpute operates on both dosage data and unphased genotype data and simultaneously imputes missing genotypes and phase at both the typed and untyped SNPs (single nucleotide polymorphisms). Finally, MendelImpute naturally extends to global and local ancestry estimation and lends itself to new strategies for data compression and hence faster data transport and sharing. Software, documentation and scripts to reproduce our results are available from https://github.com/OpenMendel/MendelImpute.jl. Supplementary data are available at Bioinformatics online.
- 21 July 2021
- Published by Oxford University Press (OUP) in Bioinformatics
- Vol. 37 (24) , 4756-4763
Open AccessA fast data-driven method for genotype imputation, phasing and local ancestry inference: MendelImpute.jl
Current methods for genotype imputation and phasing exploit the volume of data in haplotype reference panels and rely on hidden Markov models (HMMs). Existing programs all have essentially the same imputation accuracy, are computationally intensive and generally require prephasing the typed markers. We introduce a novel data-mining method for genotype imputation and phasing that substitutes highly efficient linear algebra routines for HMM calculations. This strategy, embodied in our Julia program MendelImpute.jl, avoids explicit assumptions about recombination and population structure while delivering similar prediction accuracy, better memory usage and an order of magnitude or better run-times compared to the fastest competing method. MendelImpute operates on both dosage data and unphased genotype data and simultaneously imputes missing genotypes and phase at both the typed and untyped SNPs (single nucleotide polymorphisms). Finally, MendelImpute naturally extends to global and local ancestry estimation and lends itself to new strategies for data compression and hence faster data transport and sharing. Software, documentation and scripts to reproduce our results are available from https://github.com/OpenMendel/MendelImpute.jl. Supplementary data are available at Bioinformatics online.
- Open Access
High-Performance Statistical Computing in the Computing Environments of the 2020s
Technological advances in the past decade, hardware and software alike, have made access to high-performance computing (HPC) easier than ever. We review these advances from a statistical computing perspective. Cloud computing makes access to supercomputers affordable. Deep learning software libraries make programming statistical algorithms easy and enable users to write code once and run it anywhere -- from a laptop to a workstation with multiple graphics processing units (GPUs) or a supercomputer in a cloud. Highlighting how these developments benefit statisticians, we review recent optimization algorithms that are useful for high-dimensional models and can harness the power of HPC. Code snippets are provided to demonstrate the ease of programming. We also provide an easy-to-use distributed matrix data structure suitable for HPC. Employing this data structure, we illustrate various statistical applications including large-scale positron emission tomography and $\ell_1$-regularized Cox regression. Our examples easily scale up to an 8-GPU workstation and a 720-CPU-core cluster in a cloud. As a case in point, we analyze the onset of type-2 diabetes from the UK Biobank with 200,000 subjects and about 500,000 single nucleotide polymorphisms using the HPC $\ell_1$-regularized Cox regression. Fitting this half-million-variate model takes less than 45 minutes and reconfirms known associations. To our knowledge, this is the first demonstration of the feasibility of penalized regression of survival outcomes at this scale.
Open AccessHigh-Performance Statistical Computing in the Computing Environments of the 2020s
Technological advances in the past decade, hardware and software alike, have made access to high-performance computing (HPC) easier than ever. We review these advances from a statistical computing perspective. Cloud computing makes access to supercomputers affordable. Deep learning software libraries make programming statistical algorithms easy and enable users to write code once and run it anywhere -- from a laptop to a workstation with multiple graphics processing units (GPUs) or a supercomputer in a cloud. Highlighting how these developments benefit statisticians, we review recent optimization algorithms that are useful for high-dimensional models and can harness the power of HPC. Code snippets are provided to demonstrate the ease of programming. We also provide an easy-to-use distributed matrix data structure suitable for HPC. Employing this data structure, we illustrate various statistical applications including large-scale positron emission tomography and $\ell_1$-regularized Cox regression. Our examples easily scale up to an 8-GPU workstation and a 720-CPU-core cluster in a cloud. As a case in point, we analyze the onset of type-2 diabetes from the UK Biobank with 200,000 subjects and about 500,000 single nucleotide polymorphisms using the HPC $\ell_1$-regularized Cox regression. Fitting this half-million-variate model takes less than 45 minutes and reconfirms known associations. To our knowledge, this is the first demonstration of the feasibility of penalized regression of survival outcomes at this scale.
- Open Access
- 13 July 2021
- Published by Oxford University Press (OUP) in Briefings in Bioinformatics
- Vol. 22 (6)
Computationally scalable regression modeling for ultrahigh-dimensional omics data with ParProx
Statistical analysis of ultrahigh-dimensional omics scale data has long depended on univariate hypothesis testing. With growing data features and samples, the obvious next step is to establish multivariable association analysis as a routine method to describe genotype–phenotype association. Here we present ParProx, a state-of-the-art implementation to optimize overlapping and non-overlapping group lasso regression models for time-to-event and classification analysis, with selection of variables grouped by biological priors. ParProx enables multivariable model fitting for ultrahigh-dimensional data within an architecture for parallel or distributed computing via latent variable group representation. It thereby aims to produce interpretable regression models consistent with known biological relationships among independent variables, a property often explored post hoc, not during model estimation. Simulation studies clearly demonstrate the scalability of ParProx with graphics processing units in comparison to existing implementations. We illustrate the tool using three different omics data sets featuring moderate to large numbers of variables, where we use genomic regions and biological pathways as variable groups, rendering the selected independent variables directly interpretable with respect to those groups. ParProx is applicable to a wide range of studies using ultrahigh-dimensional omics data, from genome-wide association analysis to multi-omics studies where model estimation is computationally intractable with existing implementation.
- 13 July 2021
- Published by Oxford University Press (OUP) in Briefings in Bioinformatics
- Vol. 22 (6)
Open AccessComputationally scalable regression modeling for ultrahigh-dimensional omics data with ParProx
Statistical analysis of ultrahigh-dimensional omics scale data has long depended on univariate hypothesis testing. With growing data features and samples, the obvious next step is to establish multivariable association analysis as a routine method to describe genotype–phenotype association. Here we present ParProx, a state-of-the-art implementation to optimize overlapping and non-overlapping group lasso regression models for time-to-event and classification analysis, with selection of variables grouped by biological priors. ParProx enables multivariable model fitting for ultrahigh-dimensional data within an architecture for parallel or distributed computing via latent variable group representation. It thereby aims to produce interpretable regression models consistent with known biological relationships among independent variables, a property often explored post hoc, not during model estimation. Simulation studies clearly demonstrate the scalability of ParProx with graphics processing units in comparison to existing implementations. We illustrate the tool using three different omics data sets featuring moderate to large numbers of variables, where we use genomic regions and biological pathways as variable groups, rendering the selected independent variables directly interpretable with respect to those groups. ParProx is applicable to a wide range of studies using ultrahigh-dimensional omics data, from genome-wide association analysis to multi-omics studies where model estimation is computationally intractable with existing implementation.
- Open Access
Computationally scalable regression modeling for ultrahigh-dimensional omics data with ParProx
Statistical analysis of ultrahigh-dimensional omics scale data has long depended on univariate hypothesis testing. With growing data features and samples, the obvious next step is to establish multivariable association analysis as a routine method for understanding genotype-phenotype associations. Here we present ParProx, a state-of-the-art implementation to optimize overlapping group lasso regression models for time-to-event and classification analysis, guided by biological priors through coordinated variable selection. ParProx not only enables model fitting for ultrahigh-dimensional data within the architecture for parallel or distributed computing, but also allows users to obtain interpretable regression models consistent with known biological relationships among the independent variables, a feature long neglected in statistical modeling of omics data. We demonstrate ParProx using three different omics data sets of moderate to large numbers of variables, where we use genomic regions and pathways to arrive at sparse regression models comprised of biologically related independent variables. ParProx is naturally applicable to a wide range of studies using ultrahigh-dimensional omics data, ranging from genome-wide association analysis to single cell sequencing studies where multivariable modeling is computationally intractable.
Open AccessComputationally scalable regression modeling for ultrahigh-dimensional omics data with ParProx
Statistical analysis of ultrahigh-dimensional omics scale data has long depended on univariate hypothesis testing. With growing data features and samples, the obvious next step is to establish multivariable association analysis as a routine method for understanding genotype-phenotype associations. Here we present ParProx, a state-of-the-art implementation to optimize overlapping group lasso regression models for time-to-event and classification analysis, guided by biological priors through coordinated variable selection. ParProx not only enables model fitting for ultrahigh-dimensional data within the architecture for parallel or distributed computing, but also allows users to obtain interpretable regression models consistent with known biological relationships among the independent variables, a feature long neglected in statistical modeling of omics data. We demonstrate ParProx using three different omics data sets of moderate to large numbers of variables, where we use genomic regions and pathways to arrive at sparse regression models comprised of biologically related independent variables. ParProx is naturally applicable to a wide range of studies using ultrahigh-dimensional omics data, ranging from genome-wide association analysis to single cell sequencing studies where multivariable modeling is computationally intractable.