Research Projects Directory

Scientific Approaches

We will build a cohort of All of Us participants aged 25-60 who have at least one alcohol-use survey response or EHR code from 2018-2024 and at least 12 months of clinical follow-up. Step 1: clean and merge survey alcohol measures, vital signs, and lab results; derive neighborhood deprivation indices from ZIP codes. Step 2: use machine-learning “similarity” models (graph neural networks) to infer peer ties based on shared traits and geography. Step 3: embed those ties into an open-source agent-based simulation that treats binge drinking like an infectious process (people can move from non-drinker → moderate → binge under peer influence). Each simulated agent’s cardiovascular risk will update yearly using validated CVD prediction equations fed by their own AoU blood-pressure, lipid, and HbA1c data. We will test intervention scenarios—tax increases, social-media campaigns, and combined strategies—and examine effects on CVD events and racial/ethnic risk gaps.

Scientific Approaches

Use All of Us survey data to gauge SDoH and HRQoL variables, create logistic models that take SDoH as independent variable and HRQoL as dependent variable.

Scientific Approaches

This research aims to highlight any disparities or unique trends in the prescription and effectiveness of bupropion for Hispanic populations, contributing to more personalized and equitable healthcare approaches.

Scientific Approaches

We integrate genomic and clinical data from biobank resources (e.g., All of Us, UK Biobank) to construct pan-phenome polygenic risk scores (PRS). We will utilize genome-wide association study (GWAS) summary statistics, linkage disequilibrium score regression, and multi-trait PRS methods (e.g., MTAG, PRS-CS, PRSice) to capture shared genetic architecture across phenotypes. PRS validation and evaluation of predictive performance will be done using independent datasets and cross-validation approaches. Analyses will use Python, R, PLINK, and Hail for data processing, and machine learning tools (XGBoost, scikit-learn) for optimizing predictive accuracy.

Scientific Approaches

In this study, we aim to investigate the association between genetic variants and the development of autoimmunity and cancer. We will compare the frequencies of patients with autoimmunity and cancer among subjects who harbor specific genetic variants and those who do not. We plan to use other biobanks across the globe to answer our study question more robustly.

Scientific Approaches

Our analyses will involve Cox proportional hazard modeling of strata with specific LDL and polygenic risk score ranges with the reference group being all individuals with LDL 190+ regardless of polygenic risk score. We will utilize coronary artery disease outcome data, risk factor data to use for covariate control, and genetic data to use polygenic risk scores as an independent variable (the latter of which have already been calculated and validated by our lab). Replication across other cohorts is additionally underway.

Scientific Approaches

a) Study design: nested case-control design
b) Study population:
• Case group: Adult female (≥18 years) diagnosed with lung cancer
• Control group: Adult female (≥18 years) without a history of lung cancer, matched on age, smoking status, and ethnicity.
c) Primary outcome: Incidence of lung cancer in female
d) Primary exposure:
• Reproductive factors: Age at menarche, Age at menopause, Age at first child, history of Oral Contraceptive pill, history of uterus/ovary removal, Pattern of menstruation, parity, History of Hormone replacement therapy, History of HPV vaccination
• Molecular factors: SNPs related to Estrogen Receptor (ER), Aromatase, PD-1/PDL1, Foxp3 expressing CD4+ cells/ Treg Cell, EGFR and Errb-2 gene
e) Covariates:
• Demographic:
Education, BMI, Residence, Occupation
• Clinical:
Histopathology, Site, Differentiation, Stage, Metastatic site

Scientific Approaches

Our study employs a privacy-preserving deep learning approach using the All of Us genomic dataset comprising 500 samples with 13.8M variants after quality control filtering. I'll implement a discrete diffusion probabilistic model with ancestry-conditional generation. My focus is adapting diffusion models for discrete genomic data while preserving population structure.
I'll evaluate model using: (1) statistical fidelity metrics (allele frequency distribution, linkage disequilibrium patterns, PCA overlap); (2) privacy protection measures (membership inference attack resistance, differential privacy guarantees); and (3) utility preservation metrics (GWAS reproducibility, disease risk prediction performance).
All preprocessing and model training occurs within the secured All of Us Researcher Workbench environment using Hail for genomic processing and PyTorch for model implementation.

Scientific Approaches

Python to analyze data and conduct regression analysis to find effect of bipolar drugs.

Scientific Approaches

I will conduct a retrospective observational study using the Registered Tier dataset, including survey responses, electronic health records, and physical measurements. Logistic regression models and machine learning methods such as random forests may be used to identify predictors of hip fractures. Analyses will be conducted using R and Python tools within the secure Researcher Workbench environment, following All of Us privacy and data use guidelines.

Scientific Approaches

I plan on performing analyses linking common human genetic variation to EHR phenotypes in the All of Us dataset.

Scientific Approaches

First, we will provide an overview of the different storage options within the Researcher workbench 9RW) along with their advantages and disadvantages.
Then, using the current CDR, we will query example datasets and demonstrate:
- how to save a dataframe or a plot to the RW storage
- how to read a dataframe or a plot from the storage
- how to read list data in the storage
- how to move data in/between storages.

We will use a package created by one of our DST members, Aymone Kouame, to perform these tasks in R, Python or SAS. The GitHub link to the code will also be available.

Scientific Approaches

For Educational purpose to show best practices when using jupyter notebooks for data access, storage, data manipulations - transformations, conversions, cleaning, optimization and other research support related issues that is useful for multiple AoU researchers.

Scientific Approaches

Our team will employ a multidisciplinary approach combining epidemiological methods with advanced analytics using the All of Us Researcher Workbench. We will utilize the controlled tier dataset containing EHRs, surveys, and physical measurements from a diverse participant population. We'll implement a matched cohort design, pairing IBD cases with non-IBD controls (1:3 ratio) based on age, sex, and education, followed by multivariable regression adjusting for cardiovascular risk factors, depression, and lifestyle variables. For longitudinal analysis, we'll leverage follow-up survey data where available to assess cognitive trajectory. Advanced methods will include mediation analysis to evaluate inflammatory markers (CRP, ESR) as potential pathways, and machine learning (logistic regression/Random Forest) to identify high-risk subgroups.

Scientific Approaches

Will look for genetic risk factors from genotype data. Will use Statistical and association analysis using odds ratio and fisher exact's test between risk alleles and the subject phenotype groups. Will use control data as a validation set.

Scientific Approaches

For this study, we will utilize high-performance bioinformatics tools and computational methods to efficiently process and analyze the All of Us v8 short-read Whole Genome Sequencing (srWGS) dataset. Our primary objective is to construct a Genomic Data Structure (GDS) file that includes all variants and samples, enabling efficient large-scale genomic analysis.
Datasets
We will use the All of Us v8 genotype dataset, which consists of high-coverage short-read WGS data from a diverse cohort. This dataset contains single nucleotide polymorphisms (SNPs) and small insertions/deletions (indels), and provides a unique opportunity to study population genetics and disease-associated variants in underrepresented populations.
Research Methods & Tools
Variant Data Conversion & GDS Construction
Convert Hail MatrixTable (MT) or VCF format into GDS using SeqArray (R-based tool) or Bioconductor packages.
Optimize data compression and indexing for efficient query performance.

Scientific Approaches

For this study, we will utilize high-performance bioinformatics tools and computational methods to efficiently process and analyze the All of Us v8 short-read Whole Genome Sequencing (srWGS) dataset. Our primary objective is to construct a Genomic Data Structure (GDS) file that includes all variants and samples, enabling efficient large-scale genomic analysis.
Datasets
We will use the All of Us v8 genotype dataset, which consists of high-coverage short-read WGS data from a diverse cohort. This dataset contains single nucleotide polymorphisms (SNPs) and small insertions/deletions (indels), and provides a unique opportunity to study population genetics and disease-associated variants in underrepresented populations.
Research Methods & Tools
Variant Data Conversion & GDS Construction
Convert Hail MatrixTable (MT) or VCF format into GDS using SeqArray (R-based tool) or Bioconductor packages.
Optimize data compression and indexing for efficient query performance.

Scientific Approaches

For this study, we will utilize high-performance bioinformatics tools and computational methods to efficiently process and analyze the All of Us v8 short-read Whole Genome Sequencing (srWGS) dataset. Our primary objective is to construct a Genomic Data Structure (GDS) file that includes all variants and samples, enabling efficient large-scale genomic analysis.
Datasets
We will use the All of Us v8 genotype dataset, which consists of high-coverage short-read WGS data from a diverse cohort. This dataset contains single nucleotide polymorphisms (SNPs) and small insertions/deletions (indels), and provides a unique opportunity to study population genetics and disease-associated variants in underrepresented populations.
Research Methods & Tools
Variant Data Conversion & GDS Construction
Convert Hail MatrixTable (MT) or VCF format into GDS using SeqArray (R-based tool) or Bioconductor packages.
Optimize data compression and indexing for efficient query performance.

Scientific Approaches

For this study, we will utilize high-performance bioinformatics tools and computational methods to efficiently process and analyze the All of Us v8 short-read Whole Genome Sequencing (srWGS) dataset. Our primary objective is to construct a Genomic Data Structure (GDS) file that includes all variants and samples, enabling efficient large-scale genomic analysis.
Datasets
We will use the All of Us v8 genotype dataset, which consists of high-coverage short-read WGS data from a diverse cohort. This dataset contains single nucleotide polymorphisms (SNPs) and small insertions/deletions (indels), and provides a unique opportunity to study population genetics and disease-associated variants in underrepresented populations.
Research Methods & Tools
Variant Data Conversion & GDS Construction
Convert Hail MatrixTable (MT) or VCF format into GDS using SeqArray (R-based tool) or Bioconductor packages.
Optimize data compression and indexing for efficient query performance.

Scientific Approaches

For this study, we will utilize high-performance bioinformatics tools and computational methods to efficiently process and analyze the All of Us v8 short-read Whole Genome Sequencing (srWGS) dataset. Our primary objective is to construct a Genomic Data Structure (GDS) file that includes all variants and samples, enabling efficient large-scale genomic analysis.
Datasets
We will use the All of Us v8 genotype dataset, which consists of high-coverage short-read WGS data from a diverse cohort. This dataset contains single nucleotide polymorphisms (SNPs) and small insertions/deletions (indels), and provides a unique opportunity to study population genetics and disease-associated variants in underrepresented populations.
Research Methods & Tools
Variant Data Conversion & GDS Construction
Convert Hail MatrixTable (MT) or VCF format into GDS using SeqArray (R-based tool) or Bioconductor packages.
Optimize data compression and indexing for efficient query performance.

Scientific Approaches

For this study, we will utilize high-performance bioinformatics tools and computational methods to efficiently process and analyze the All of Us v8 short-read Whole Genome Sequencing (srWGS) dataset. Our primary objective is to construct a Genomic Data Structure (GDS) file that includes all variants and samples, enabling efficient large-scale genomic analysis.
Datasets
We will use the All of Us v8 genotype dataset, which consists of high-coverage short-read WGS data from a diverse cohort. This dataset contains single nucleotide polymorphisms (SNPs) and small insertions/deletions (indels), and provides a unique opportunity to study population genetics and disease-associated variants in underrepresented populations.
Research Methods & Tools
Variant Data Conversion & GDS Construction
Convert Hail MatrixTable (MT) or VCF format into GDS using SeqArray (R-based tool) or Bioconductor packages.
Optimize data compression and indexing for efficient query performance.

Scientific Approaches

For this study, we will utilize high-performance bioinformatics tools and computational methods to efficiently process and analyze the All of Us v8 short-read Whole Genome Sequencing (srWGS) dataset. Our primary objective is to construct a Genomic Data Structure (GDS) file that includes all variants and samples, enabling efficient large-scale genomic analysis.
Datasets
We will use the All of Us v8 genotype dataset, which consists of high-coverage short-read WGS data from a diverse cohort. This dataset contains single nucleotide polymorphisms (SNPs) and small insertions/deletions (indels), and provides a unique opportunity to study population genetics and disease-associated variants in underrepresented populations.
Research Methods & Tools
Variant Data Conversion & GDS Construction
Convert Hail MatrixTable (MT) or VCF format into GDS using SeqArray (R-based tool) or Bioconductor packages.
Optimize data compression and indexing for efficient query performance.

Scientific Approaches

For this study, we will utilize high-performance bioinformatics tools and computational methods to efficiently process and analyze the All of Us v8 short-read Whole Genome Sequencing (srWGS) dataset. Our primary objective is to construct a Genomic Data Structure (GDS) file that includes all variants and samples, enabling efficient large-scale genomic analysis.
Datasets
We will use the All of Us v8 genotype dataset, which consists of high-coverage short-read WGS data from a diverse cohort. This dataset contains single nucleotide polymorphisms (SNPs) and small insertions/deletions (indels), and provides a unique opportunity to study population genetics and disease-associated variants in underrepresented populations.
Research Methods & Tools
Variant Data Conversion & GDS Construction
Convert Hail MatrixTable (MT) or VCF format into GDS using SeqArray (R-based tool) or Bioconductor packages.
Optimize data compression and indexing for efficient query performance.

Scientific Approaches

For this study, we will utilize high-performance bioinformatics tools and computational methods to efficiently process and analyze the All of Us v8 short-read Whole Genome Sequencing (srWGS) dataset. Our primary objective is to construct a Genomic Data Structure (GDS) file that includes all variants and samples, enabling efficient large-scale genomic analysis.
Datasets
We will use the All of Us v8 genotype dataset, which consists of high-coverage short-read WGS data from a diverse cohort. This dataset contains single nucleotide polymorphisms (SNPs) and small insertions/deletions (indels), and provides a unique opportunity to study population genetics and disease-associated variants in underrepresented populations.
Research Methods & Tools
Variant Data Conversion & GDS Construction
Convert Hail MatrixTable (MT) or VCF format into GDS using SeqArray (R-based tool) or Bioconductor packages.
Optimize data compression and indexing for efficient query performance.

Scientific Approaches

For this study, we will utilize high-performance bioinformatics tools and computational methods to efficiently process and analyze the All of Us v8 short-read Whole Genome Sequencing (srWGS) dataset. Our primary objective is to construct a Genomic Data Structure (GDS) file that includes all variants and samples, enabling efficient large-scale genomic analysis.
Datasets
We will use the All of Us v8 genotype dataset, which consists of high-coverage short-read WGS data from a diverse cohort. This dataset contains single nucleotide polymorphisms (SNPs) and small insertions/deletions (indels), and provides a unique opportunity to study population genetics and disease-associated variants in underrepresented populations.
Research Methods & Tools
Variant Data Conversion & GDS Construction
Convert Hail MatrixTable (MT) or VCF format into GDS using SeqArray (R-based tool) or Bioconductor packages.
Optimize data compression and indexing for efficient query performance.