Ulzee An
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
Ulzee An has a total of 64 co-authors. Here are the top co-authors by number of shared publications:
- Sriram Sankararaman
Department of Computer Science, UCLA, Los Angeles, CA, USA
6 shared publications - Jonathan Flint
Department of Human Genetics, David Geffen School of Medicine, University of California, Los Angeles, Los Angeles, USA
4 shared publications - Kenneth S. Kendler
Virginia Institute for Psychiatric and Behavioral Genetics, Department of Psychiatry Virginia Commonwealth University Richmond Virginia USA
4 shared publications - Na Cai
Department of Biosystems Science and Engineering, ETH Zurich, Basel, Switzerland
4 shared publications - Silviu-Alin Bacanu
Department of Psychiatry, Virginia Commonwealth University
4 shared publications - Andy Dahl
Section of Genetic Medicine, University of Chicago, Chicago, USA
3 shared publications
Latest Publications
- Open Access
Genome-wide LD scores enable robust heritability and genetic correlation estimation from summary statistics across genetic ancestries
Our knowledge of human complex trait architecture across diverse populations remains limited. Estimating SNP heritability and genetic correlation from GWAS summary statistics – successful in revealing genetic architecture in homogeneous cohorts – is challenging in structured and admixed populations due, in part, to the difficulty of representing long-range LD. We present SUMMIT, a statistical framework that combines genome-wide LD scores, efficiently estimated using random projections, with GWAS summary statistics to estimate genome-wide and annotation-specific heritability and genetic correlation. In simulations that explore diverse aspects of genetic architecture across European (EUR), South Asian (SAS), and admixed African-European cohorts (AFR) in the UK Biobank (UKB), SUMMIT had increased accuracy and reduced bias compared to prior methods, with the largest gains in non-European ancestries reflecting improved modeling of long-range LD. Across 40 UKB traits, SUMMIT produced estimates of genome-wide common-SNP heritability and heritability enrichments in genomic annotations that were broadly concordant across ancestries with the SUMMIT estimates more closely matching recent whole-genome sequencing based estimates. SUMMIT substantially improved genetic correlation analysis recovering more significant findings across ancestries (23 SAS and 3 AFR at FDR < 0.05 compared with none and two for cov-LDSC) with the non-European discoveries showing concordant effects in Europeans suggesting that, like heritability, global patterns of pleiotropy are largely shared across ancestries. By combining accurate genome-wide LD and efficient moment estimation, SUMMIT enables investigation into how heritability and genetic correlation vary across populations and genomic annotations, including in structured and admixed cohorts, at the genome-wide, annotation-specific, and cross-trait levels.
Open AccessGenome-wide LD scores enable robust heritability and genetic correlation estimation from summary statistics across genetic ancestries
Our knowledge of human complex trait architecture across diverse populations remains limited. Estimating SNP heritability and genetic correlation from GWAS summary statistics – successful in revealing genetic architecture in homogeneous cohorts – is challenging in structured and admixed populations due, in part, to the difficulty of representing long-range LD. We present SUMMIT, a statistical framework that combines genome-wide LD scores, efficiently estimated using random projections, with GWAS summary statistics to estimate genome-wide and annotation-specific heritability and genetic correlation. In simulations that explore diverse aspects of genetic architecture across European (EUR), South Asian (SAS), and admixed African-European cohorts (AFR) in the UK Biobank (UKB), SUMMIT had increased accuracy and reduced bias compared to prior methods, with the largest gains in non-European ancestries reflecting improved modeling of long-range LD. Across 40 UKB traits, SUMMIT produced estimates of genome-wide common-SNP heritability and heritability enrichments in genomic annotations that were broadly concordant across ancestries with the SUMMIT estimates more closely matching recent whole-genome sequencing based estimates. SUMMIT substantially improved genetic correlation analysis recovering more significant findings across ancestries (23 SAS and 3 AFR at FDR < 0.05 compared with none and two for cov-LDSC) with the non-European discoveries showing concordant effects in Europeans suggesting that, like heritability, global patterns of pleiotropy are largely shared across ancestries. By combining accurate genome-wide LD and efficient moment estimation, SUMMIT enables investigation into how heritability and genetic correlation vary across populations and genomic annotations, including in structured and admixed cohorts, at the genome-wide, annotation-specific, and cross-trait levels.
- Open Access
Raptor: Scalable Train-Free Embeddings for 3D Medical Volumes Leveraging Pretrained 2D Foundation Models
Current challenges in developing foundational models for volumetric imaging data, such as magnetic resonance imaging (MRI), stem from the computational complexity of training state-of-the-art architectures in high dimensions and curating sufficiently large datasets of volumes. To address these challenges, we introduce Raptor (Random Planar Tensor Reduction), a train-free method for generating semantically rich embeddings for volumetric data. Raptor leverages a frozen 2D foundation model, pretrained on natural images, to extract visual tokens from individual cross-sections of medical volumes. These tokens are then spatially compressed using random projections, significantly reducing computational complexity while retaining semantic information. Extensive experiments on ten diverse medical volume tasks verify the superior performance of Raptor over state-of-the-art methods, including those pretrained exclusively on medical volumes (+3% SuPreM, +6% MISFM, +10% Merlin, +13% VoCo, and +14% SLIViT), while entirely bypassing the need for costly training. Our results highlight the effectiveness and versatility of Raptor as a foundation for advancing deep learning-based methods for medical volumes.
Open AccessRaptor: Scalable Train-Free Embeddings for 3D Medical Volumes Leveraging Pretrained 2D Foundation Models
Current challenges in developing foundational models for volumetric imaging data, such as magnetic resonance imaging (MRI), stem from the computational complexity of training state-of-the-art architectures in high dimensions and curating sufficiently large datasets of volumes. To address these challenges, we introduce Raptor (Random Planar Tensor Reduction), a train-free method for generating semantically rich embeddings for volumetric data. Raptor leverages a frozen 2D foundation model, pretrained on natural images, to extract visual tokens from individual cross-sections of medical volumes. These tokens are then spatially compressed using random projections, significantly reducing computational complexity while retaining semantic information. Extensive experiments on ten diverse medical volume tasks verify the superior performance of Raptor over state-of-the-art methods, including those pretrained exclusively on medical volumes (+3% SuPreM, +6% MISFM, +10% Merlin, +13% VoCo, and +14% SLIViT), while entirely bypassing the need for costly training. Our results highlight the effectiveness and versatility of Raptor as a foundation for advancing deep learning-based methods for medical volumes.
- 1 October 2024
- Published by Springer Nature in Nature Biomedical Engineering
- Vol. 9 (4) , 507-520
Accurate prediction of disease-risk factors from volumetric medical scans by a deep vision model pre-trained with 2D scans
The application of machine learning to tasks involving volumetric biomedical imaging is constrained by the limited availability of annotated datasets of three-dimensional (3D) scans for model training. Here we report a deep-learning model pre-trained on 2D scans (for which annotated data are relatively abundant) that accurately predicts disease-risk factors from 3D medical-scan modalities. The model, which we named SLIViT (for ‘slice integration by vision transformer’), preprocesses a given volumetric scan into 2D images, extracts their feature map and integrates it into a single prediction. We evaluated the model in eight different learning tasks, including classification and regression for six datasets involving four volumetric imaging modalities (computed tomography, magnetic resonance imaging, optical coherence tomography and ultrasound). SLIViT consistently outperformed domain-specific state-of-the-art models and was typically as accurate as clinical specialists who had spent considerable time manually annotating the analysed scans. Automating diagnosis tasks involving volumetric scans may save valuable clinician hours, reduce data acquisition costs and duration, and help expedite medical research and clinical applications.
- 1 October 2024
- Published by Springer Nature in Nature Biomedical Engineering
- Vol. 9 (4) , 507-520
Accurate prediction of disease-risk factors from volumetric medical scans by a deep vision model pre-trained with 2D scans
The application of machine learning to tasks involving volumetric biomedical imaging is constrained by the limited availability of annotated datasets of three-dimensional (3D) scans for model training. Here we report a deep-learning model pre-trained on 2D scans (for which annotated data are relatively abundant) that accurately predicts disease-risk factors from 3D medical-scan modalities. The model, which we named SLIViT (for ‘slice integration by vision transformer’), preprocesses a given volumetric scan into 2D images, extracts their feature map and integrates it into a single prediction. We evaluated the model in eight different learning tasks, including classification and regression for six datasets involving four volumetric imaging modalities (computed tomography, magnetic resonance imaging, optical coherence tomography and ultrasound). SLIViT consistently outperformed domain-specific state-of-the-art models and was typically as accurate as clinical specialists who had spent considerable time manually annotating the analysed scans. Automating diagnosis tasks involving volumetric scans may save valuable clinician hours, reduce data acquisition costs and duration, and help expedite medical research and clinical applications.
- Open Access
- 20 November 2023
- Published by Springer Nature in Nature Genetics
- Vol. 55 (12) , 2269-2276
Deep learning-based phenotype imputation on population-scale biobank data increases genetic discoveries
Biobanks that collect deep phenotypic and genomic data across many individuals have emerged as a key resource in human genetics. However, phenotypes in biobanks are often missing across many individuals, limiting their utility. We propose AutoComplete, a deep learning-based imputation method to impute or ‘fill-in’ missing phenotypes in population-scale biobank datasets. When applied to collections of phenotypes measured across ~300,000 individuals from the UK Biobank, AutoComplete substantially improved imputation accuracy over existing methods. On three traits with notable amounts of missingness, we show that AutoComplete yields imputed phenotypes that are genetically similar to the originally observed phenotypes while increasing the effective sample size by about twofold on average. Further, genome-wide association analyses on the resulting imputed phenotypes led to a substantial increase in the number of associated loci. Our results demonstrate the utility of deep learning-based phenotype imputation to increase power for genetic discoveries in existing biobank datasets.
- 20 November 2023
- Published by Springer Nature in Nature Genetics
- Vol. 55 (12) , 2269-2276
Open AccessDeep learning-based phenotype imputation on population-scale biobank data increases genetic discoveries
Biobanks that collect deep phenotypic and genomic data across many individuals have emerged as a key resource in human genetics. However, phenotypes in biobanks are often missing across many individuals, limiting their utility. We propose AutoComplete, a deep learning-based imputation method to impute or ‘fill-in’ missing phenotypes in population-scale biobank datasets. When applied to collections of phenotypes measured across ~300,000 individuals from the UK Biobank, AutoComplete substantially improved imputation accuracy over existing methods. On three traits with notable amounts of missingness, we show that AutoComplete yields imputed phenotypes that are genetically similar to the originally observed phenotypes while increasing the effective sample size by about twofold on average. Further, genome-wide association analyses on the resulting imputed phenotypes led to a substantial increase in the number of associated loci. Our results demonstrate the utility of deep learning-based phenotype imputation to increase power for genetic discoveries in existing biobank datasets.
- Open Access
- 20 November 2023
- Published by Springer Nature in Nature Genetics
- Vol. 55 (12) , 2082-2093
Phenotype integration improves power and preserves specificity in biobank-based genetic studies of major depressive disorder
Biobanks often contain several phenotypes relevant to diseases such as major depressive disorder (MDD), with partly distinct genetic architectures. Researchers face complex tradeoffs between shallow (large sample size, low specificity/sensitivity) and deep (small sample size, high specificity/sensitivity) phenotypes, and the optimal choices are often unclear. Here we propose to integrate these phenotypes to combine the benefits of each. We use phenotype imputation to integrate information across hundreds of MDD-relevant phenotypes, which significantly increases genome-wide association study (GWAS) power and polygenic risk score (PRS) prediction accuracy of the deepest available MDD phenotype in UK Biobank, LifetimeMDD. We demonstrate that imputation preserves specificity in its genetic architecture using a novel PRS-based pleiotropy metric. We further find that integration via summary statistics also enhances GWAS power and PRS predictions, but can introduce nonspecific genetic effects depending on input. Our work provides a simple and scalable approach to improve genetic studies in large biobanks by integrating shallow and deep phenotypes.
- 20 November 2023
- Published by Springer Nature in Nature Genetics
- Vol. 55 (12) , 2082-2093
Open AccessPhenotype integration improves power and preserves specificity in biobank-based genetic studies of major depressive disorder
Biobanks often contain several phenotypes relevant to diseases such as major depressive disorder (MDD), with partly distinct genetic architectures. Researchers face complex tradeoffs between shallow (large sample size, low specificity/sensitivity) and deep (small sample size, high specificity/sensitivity) phenotypes, and the optimal choices are often unclear. Here we propose to integrate these phenotypes to combine the benefits of each. We use phenotype imputation to integrate information across hundreds of MDD-relevant phenotypes, which significantly increases genome-wide association study (GWAS) power and polygenic risk score (PRS) prediction accuracy of the deepest available MDD phenotype in UK Biobank, LifetimeMDD. We demonstrate that imputation preserves specificity in its genetic architecture using a novel PRS-based pleiotropy metric. We further find that integration via summary statistics also enhances GWAS power and PRS predictions, but can introduce nonspecific genetic effects depending on input. Our work provides a simple and scalable approach to improve genetic studies in large biobanks by integrating shallow and deep phenotypes.
- Open Access
Phenotype integration improves power and preserves specificity in biobank-based genetic studies of MDD
Biobanks often contain several phenotypes relevant to a given disorder, and researchers face complex tradeoffs between shallow phenotypes (high sample size, low specificity and sensitivity) and deep phenotypes (low sample size, high specificity and sensitivity). Here, we study an extreme case: Major Depressive Disorder (MDD) in UK Biobank. Previous studies found that shallow and deep MDD phenotypes have qualitatively distinct genetic architectures, but it remains unclear which are optimal for scientific study or clinical prediction. We propose a new framework to get the best of both worlds by integrating together information across hundreds of MDD-relevant phenotypes. First, we use phenotype imputation to increase sample size for the deepest available MDD phenotype, which dramatically improves GWAS power (increases #loci ~10 fold) and PRS accuracy (increases R2 ~2 fold). Further, we show the genetic architecture of the imputed phenotype remains specific to MDD using genetic correlation, PRS prediction in external clinical cohorts, and a novel PRS-based pleiotropy metric. We also develop a complementary approach to improve specificity of GWAS on shallow MDD phenotypes by adjusting for phenome-wide PCs. Finally, we study phenotype integration at the level of GWAS summary statistics, which can increase GWAS and PRS power but introduces non-MDD-specific signals. Our work provides a simple and scalable recipe to improve genetic studies in large biobanks by combining the sample size of shallow phenotypes with the sensitivity and specificity of deep phenotypes.
Open AccessPhenotype integration improves power and preserves specificity in biobank-based genetic studies of MDD
Biobanks often contain several phenotypes relevant to a given disorder, and researchers face complex tradeoffs between shallow phenotypes (high sample size, low specificity and sensitivity) and deep phenotypes (low sample size, high specificity and sensitivity). Here, we study an extreme case: Major Depressive Disorder (MDD) in UK Biobank. Previous studies found that shallow and deep MDD phenotypes have qualitatively distinct genetic architectures, but it remains unclear which are optimal for scientific study or clinical prediction. We propose a new framework to get the best of both worlds by integrating together information across hundreds of MDD-relevant phenotypes. First, we use phenotype imputation to increase sample size for the deepest available MDD phenotype, which dramatically improves GWAS power (increases #loci ~10 fold) and PRS accuracy (increases R2 ~2 fold). Further, we show the genetic architecture of the imputed phenotype remains specific to MDD using genetic correlation, PRS prediction in external clinical cohorts, and a novel PRS-based pleiotropy metric. We also develop a complementary approach to improve specificity of GWAS on shallow MDD phenotypes by adjusting for phenome-wide PCs. Finally, we study phenotype integration at the level of GWAS summary statistics, which can increase GWAS and PRS power but introduces non-MDD-specific signals. Our work provides a simple and scalable recipe to improve genetic studies in large biobanks by combining the sample size of shallow phenotypes with the sensitivity and specificity of deep phenotypes.
- Open Access
Deep Learning-based Phenotype Imputation on Population-scale Biobank Data Increases Genetic Discoveries
Biobanks that collect deep phenotypic and genomic data across large numbers of individuals have emerged as a key resource for human genetic research. However, phenotypes acquired as part of Biobanks are often missing across many individuals, limiting the utility of these datasets. The ability to accurately impute or “fill-in” missing phenotypes is critical to harness the power of population-scale Biobank datasets. We propose AutoComplete, a deep learning-based imputation method which can accurately impute missing phenotypes in population-scale Biobank datasets. When applied to collections of phenotypes measured across ≈ 300K individuals from the UK Biobank, AutoComplete improved imputation accuracy over existing 2 methods (average improvement in r 2 of 18% for all phenotypes and 42% for binary phenotypes). We explored the utility of phenotype imputation for improving the power of genome-wide association studies (GWAS) by applying our method to a group of five clinically relevant traits with an average missigness rate of 83% (67% to 94%) leading to an an increase in effective sample size of ≈2-fold on average (0.5 to 3.3-fold across the phenotypes). GWAS on the resulting imputed phenotypes led to an increase in the total number of loci significantly associated to the traits from four to 129. Our results demonstrate the utility of deep-learning based imputation to increase power for genetic discoveries in existing biobank data sets.
Open AccessDeep Learning-based Phenotype Imputation on Population-scale Biobank Data Increases Genetic Discoveries
Biobanks that collect deep phenotypic and genomic data across large numbers of individuals have emerged as a key resource for human genetic research. However, phenotypes acquired as part of Biobanks are often missing across many individuals, limiting the utility of these datasets. The ability to accurately impute or “fill-in” missing phenotypes is critical to harness the power of population-scale Biobank datasets. We propose AutoComplete, a deep learning-based imputation method which can accurately impute missing phenotypes in population-scale Biobank datasets. When applied to collections of phenotypes measured across ≈ 300K individuals from the UK Biobank, AutoComplete improved imputation accuracy over existing 2 methods (average improvement in r 2 of 18% for all phenotypes and 42% for binary phenotypes). We explored the utility of phenotype imputation for improving the power of genome-wide association studies (GWAS) by applying our method to a group of five clinically relevant traits with an average missigness rate of 83% (67% to 94%) leading to an an increase in effective sample size of ≈2-fold on average (0.5 to 3.3-fold across the phenotypes). GWAS on the resulting imputed phenotypes led to an increase in the total number of loci significantly associated to the traits from four to 129. Our results demonstrate the utility of deep-learning based imputation to increase power for genetic discoveries in existing biobank data sets.
- 29 April 2022
- Published by Springer Nature
- p. 385-386
AutoComplete: Deep Learning-Based Phenotype Imputation for Large-Scale Biomedical Data
- 29 April 2022
- Published by Springer Nature
- p. 385-386
AutoComplete: Deep Learning-Based Phenotype Imputation for Large-Scale Biomedical Data
- Vol. 267, 1462-1482
Raptor: Scalable Train-Free Embeddings for 3D Medical Volumes Leveraging Pretrained 2D Foundation Models
Current challenges in developing foundational models for volumetric imaging data, such as magnetic resonance imaging (MRI), stem from the computational complexity of training state-of-the-art architectures in high dimensions and curating sufficiently large datasets of volumes. To address these challenges, we introduce Raptor (Random Planar Tensor Reduction), a train-free method for generating semantically rich embeddings for volumetric data. Raptor leverages a frozen 2D foundation model, pretrained on natural images, to extract visual tokens from individual cross-sections of medical volumes. These tokens are then spatially compressed using random projections, significantly reducing computational complexity while retaining semantic information. Extensive experiments on ten diverse medical volume tasks verify the superior performance of Raptor over state-of-the-art methods, including those pretrained exclusively on medical volumes (+3% SuPreM, +6% MISFM, +10% Merlin, +13% VoCo, and +14% SLIViT), while entirely bypassing the need for costly training. Our results highlight the effectiveness and versatility of Raptor as a foundation for advancing deep learning-based methods for medical volumes (code: github.com/sriramlab/raptor).
- Vol. 267, 1462-1482
Raptor: Scalable Train-Free Embeddings for 3D Medical Volumes Leveraging Pretrained 2D Foundation Models
Current challenges in developing foundational models for volumetric imaging data, such as magnetic resonance imaging (MRI), stem from the computational complexity of training state-of-the-art architectures in high dimensions and curating sufficiently large datasets of volumes. To address these challenges, we introduce Raptor (Random Planar Tensor Reduction), a train-free method for generating semantically rich embeddings for volumetric data. Raptor leverages a frozen 2D foundation model, pretrained on natural images, to extract visual tokens from individual cross-sections of medical volumes. These tokens are then spatially compressed using random projections, significantly reducing computational complexity while retaining semantic information. Extensive experiments on ten diverse medical volume tasks verify the superior performance of Raptor over state-of-the-art methods, including those pretrained exclusively on medical volumes (+3% SuPreM, +6% MISFM, +10% Merlin, +13% VoCo, and +14% SLIViT), while entirely bypassing the need for costly training. Our results highlight the effectiveness and versatility of Raptor as a foundation for advancing deep learning-based methods for medical volumes (code: github.com/sriramlab/raptor).