Research Projects Directory

Scientific Approaches

We would like to use data on patients diagnosed with uveitis and examine variables such as number of steroid injections and rates of management with DMARDs and biologics. We will perform logistic/linear regressions according to the data to examine associations between SES factors and uveitis outcomes and management.

Scientific Approaches

My approach is grounded in NIH research that shows a link between lower resting heart rate and better cardiovascular health and longevity. I will extract Fitbit data from the All of Us Curated Data Repository (CDR) for a cohort of participants. After cleaning the Fitbit datasets, I will conduct exploratory data analysis. For data preprocessing, I will create the target variable (i.e., resting heart rate) and perform downsampling. To develop a robust model for predicting resting heart rate based on daily summary statistics related to sleep and physical activity, I will utilize two machine learning algorithms: Support Vector Regression (SVR) and Random Forest. I will also implement feature engineering and hyperparameter tuning to improve model performance. Finally, I will identify key features of daily sleep and physical activity that correlate with minimal resting heart rate.

Scientific Approaches

We plan to use Python and SQL through Jupyter Notebook to analyze the data set and explore the relationships between other diseases.

Scientific Approaches

People 18 years and older who have answered survey questions about mental health symptoms, diagnoses, and experiences of discrimination will be included in this study. Participants will be categorized based on the severity of mental illness symptoms (normal, mild, moderate, or severe) and will also be classified based on their diagnosis(es).

Descriptive statistics (cross-tabulations with chi-squared tests; means with t-tests) will be used to assess the sociodemographic characteristics of individuals with mental health issues based on symptoms reported and also done separately based on the reported mental health diagnosis(es). Bivariate regression analyses will be performed to analyze the predictors of an individual endorsing a1) mental illness symptoms, a2) mental health diagnoses, and b) discrimination/stigma to get a sense of the data and covariates.

Scientific Approaches

I will be using survey data, electronic health records, and demographic information. Descriptive Statistics, data visualization, predictive modeling.

Scientific Approaches

How are social needs related to unmet mental health need?
• One of many independent variables: transportation need
o Other IVs: insurance, food insecurity, social support, social isolation, racial discrimination, etc.
• Dependent Variable: unmet mental health need

Scientific Approaches

We will select 20,000 unrelated participants (17.5% from each of AFR, ASN, AMR, EUR and 30% from OTH) from this cohort based on age, sex, availability of EHR
data and genetic ancestry. We will randomly sample training sets (N=2000) from the participants using naive sampling as well as sampling based on genetic representation of the cohort. We will calculate the PRS for 5 traits and then standardize them using methods used previously (doi:10.1002/gepi.22590). We will compare the expected number of participants in the upper tails, by genetic ancestry, to the observed counts, to determine which method(s) give the best standardization. We will tabulate the performance of each method by training data used, heritability (h^2) and number of SNPs for each condition: Breast Cancer (h^2=0.3 313 SNPs), Coronary Heart Disease (h^2=0.22; 5.4e5 SNPs), Hypercholesterolemia (h^2=0.56; 9k SNPs), Prostate Cancer (h^2=0.58; 269 SNPs), Type 2 diabetes (h^2=0.15, 1e6 SNPs).

Scientific Approaches

We will assess how social health factors influence the severity rare mendelian diseases. We will assess whether common diseases are moderate forms of severe rare mendelian diseases. We will examine the genetic basis of these rare and common disease is influenced by loss of function and missense variants.

Scientific Approaches

To calculate mosaic loss, the study will use whole genome sequence data of participants to quantify chromosome X dosage across individuals of different ages and ancestries. Pending the results of these findings, the study will use phenotypic data from the All of Us biobank, leveraging genome-wide association studies (GWAS) to understand phenotypic impacts of mosaic chromosome X loss.

Scientific Approaches

We will use various machine learning approaches (e.g., deep learning, foundation models) to identify patients at risk for opioid use disorder. This will involve creating a cohort of all patients prescribed an opioid during their case. The population will be split into those who had or did not have a diagnosis of opioid use disorder (e.g., ICD10 F11.xx). Predictor variables that will be included are responses to survey questions (e.g., social determinants of health), demographic/geographic data, diagnosis codes, procedure codes, medications, and genomic information. These models will include genomic information from SNPs as well as markers discovered via GWAS. We will train models with a portion of the dataset and will validate the models on a separate test set.

Scientific Approaches

We will create workflows that align, intersect, extract, integrate, and analyze known predispositions of heart maladies mutations in the All of Us cohort.
Align: We will use the UCSC genome browser tool to ensure that predisposition variants match the human reference build of All of Us.
Intersect: We will use “bedtools intersect” and “BigQuery” to identify predisposition variants in whole genome sequencing mutation files (VCFs).
Extract: We will store all suspected predisposition variants as a first “data freeze”. This dataset will be our “training-set”. We will use subsequent All of Us data releases as “test-sets” for any novel associations or statical models we identify.
Integrate: Using genomics data and insurance billing codes, we will visualize the relationships between predisposition variants, cancer occurrences, and other reported health problems.
Analyze: We will build custom scripts in Python and R to identify associations found when combining genomics and phenotypic data.

Scientific Approaches

We will use the All of Us Research Program dataset to analyze demographic, clinical, and treatment data for Latinx individuals diagnosed with rosacea. Our study will employ a mixed-methods approach: quantitative analysis to assess the prevalence of advanced rosacea treatments (e.g., laser therapy, compounded medications) and qualitative analysis to explore barriers to accessing these treatments. Statistical methods, including logistic regression, will be used to identify predictors of advanced treatment use. We will also perform subgroup analyses to evaluate treatment outcomes by factors such as socioeconomic status, language preference, and geographic location. Tools such as R and Python will be used for data cleaning and statistical modeling, and NVivo will assist in analyzing qualitative data from survey responses if applicable. This approach will provide a comprehensive understanding of treatment disparities and inform targeted interventions to address them.

Scientific Approaches

Create a cohort of patients who underwent breast reconstruction using autologous tissue and then had failure of the flap to investigate if they had pro-thrombotic genes.

Scientific Approaches

We aim to structure this study as a retrospective analysis of patients with salivary stones matched to patients without salivary stones. We will collect patient demographic data as well as condition data about hypertension, diabetes, alcohol use, obesity, and others. We will then perform a logistic regression to identify the effect size of these different condition and demographic variables in predicting salivary stones. We will optimize the logistic regression with various machine learning packages in python.

Scientific Approaches

We plan to evaluate pre-existing summary statistics assessing single variant and rare variant analyses of the All of Us cohort. We also propose to access genotype data derived from sequence data as well as phenotype data to allow for definition of case, control and quantitative trait values, as well as demographic or covariate values such as age, sex, BMI, any relevant experimental batches or similar covariates, other genetic covariates, and/or principal components to allow for correction of fine-scale ancestry. We will perform association analyses using Regenie, Remeta, Metal or similar software. We plan to perform association tests for case-control status and quantitative traits with single variants, as well as aggregated variants (e.g. rare variant burden tests), using direct genotypes, imputed variants, and may use approximate correction methods that use LD.

Scientific Approaches

To achieve this, we will categorize both rare and common genetic variants within established biological pathways, utilizing publicly available functional genomics datasets. We will employ gene expression of risk loci to identify potential target genes that co-locate with SLE genome-wide association study (GWAS) hits.

Scientific Approaches

Collect persons from variants of ZC3H12A. From those persons then extract out phenotypes and track novel phenotypes or any phenotypes, and run genomic and phenotypic analyses.

Scientific Approaches

We will be focusing on creating a diverse dataset in our investigation. Our research methods consist of creating a dataset, and further analyzing the data.

Scientific Approaches

Not applicable - these notebooks demonstrate example analysis how to use Hail and PLINK to perform genome-wide association studies using the All of Us genomic data and phenotypic data.

Scientific Approaches

We plan to evaluate pre-existing summary statistics assessing single variant and rare variant analyses of the All of Us cohort.  We also propose to access genotype data derived from sequence data as well as phenotype data to allow for definition of case, control and quantitative trait values, as well as demographic or covariate values such as age, sex, BMI, any relevant experimental batches or similar covariates, other genetic covariates, and/or principal components to allow for correction of fine-scale ancestry.  We will perform association analyses using Regenie, Remeta, Metal or similar software.  We plan to perform association tests for case-control status and quantitative traits with single variants, as well as aggregated variants (e.g. rare variant burden tests), using direct genotypes, imputed variants, and may use approximate correction methods that use LD.

Scientific Approaches

We will predict several common outcomes used by payers and hospitals for population health management services including use of the ED and hospital (re)admissions using demographic, claims, and EHR data. We will assess the performance of the algorithms by varies definitions of patient access using patient survey data on health care use and access.

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 research will employ a multi-staged analytical approach combining genetic and clinical data analysis. Initially, our focus will be on recapitulating the scientific results from the latest set of scientific publications on CAD-specific PRS using established genome-wide association study (GWAS) summary statistics and validated methodologies.

Next, we hope to expand on the published PRS into disease progression analysis. We will define objective endpoints including: progression to severe CAD (determined by coronary intervention requirements) and major adverse cardiovascular events. We still intend to utilize standard methods for the PRS calculations, incorporating the most recent meta-analyses of CAD-associated variants. We will assess PRS prediction performance through calibration plots and area under the receiver operating characteristic curve (AUC-ROC).

Scientific Approaches

We will create three data sets for analysis:

1. A list of diseases rated in the following ways:

a. Evidence Class
i. Strong evidence it is autoimmune
ii. Moderate evidence it is autoimmune
iii. Weak evidence for autoimmunity
iv. A comorbidity of autoimmune disease
v. Symptom or symptom set with no known mechanism

b. Autoinflammatory versus autoimmune flag

c. “Not always autoimmune” flag – to indicate diseases that could have alternative mechanisms of cause

2. A list of patients, anonymized, with socioeconomic, geographic and other data that would be of interest to patients and public health officials to understand which communities are affected by these diseases
3. Outcomes data for patients over time assessing quality of life using PROMIS metrics
4. We will develop statistics analyzing the association of variants known to affect autoimmune diseases for specific diseases to see if those variants corelate with other autoimmune diseases.

Scientific Approaches

We will create three data sets for analysis:

1. A list of diseases rated in the following ways:

a. Evidence Class
i. Strong evidence it is autoimmune
ii. Moderate evidence it is autoimmune
iii. Weak evidence for autoimmunity
iv. A comorbidity of autoimmune disease
v. Symptom or symptom set with no known mechanism

b. Autoinflammatory versus autoimmune flag

c. “Not always autoimmune” flag – to indicate diseases that could have alternative mechanisms of cause

2. A list of patients, anonymized, with socioeconomic, geographic and other data that would be of interest to patients and public health officials to understand which communities are affected by these diseases
3. Outcomes data for patients over time assessing quality of life using PROMIS metrics