INTEGRATING MULTI-MODAL DATA WITH ELECTRONIC HEALTH RECORDS
We seek to integrate multiple layers of high-dimensional multi-modal data including genomics, transcriptomics, copy number variation, neuroimaging, lifestyle-related factors, and social determinants of health with electronic health records to understand the interplay between different layers of data and their role on diseases. While gene expression data is typically measured on a smaller sample of individuals, rapid advances in genotyping technologies have enabled genetic and phenotypic data to be available on a much larger sample size, typically from large scale medical biobanks or other prospective cohorts. Our lab has expertise in application of statistical and computational methods that can integrate multi-omic data from non-overlapping samples to (a) better characterize genetic variants (common, rare and structural) underlying these complex phenotypes as well as (b) understand shared genetic interrelationships among these phenotypes. We are also developing statistical methods that can integrate neuroimaging data with genomics to predict risk of cognitive decline and associated comorbidities.
UNDERSTANDING DIFFERENCES IN DISEASE RISK BETWEEN POPULATION SUBGROUPS
We are interested in studying factors that contribute to an increase in risk of complex human diseases. One of the conditions of current interest to us is late-age cognitive decline that is a precursor to dementia. There are several multimorbidities (e.g. heart disease, type II diabetes) that increase the risk of cognitive decline, whose incidence rates tend to differ by population subgroups e.g., ethnic backgrounds, sex, rural-urban status. We have developed Bayesian genotype-by-subgroup interaction models that can quantify for any given trait, the extent of heterogeneity between these subgroups that is attributable to genetic variation. We are interested in leveraging longitudinal EHR of patients to incorporate other risk factors (e.g. social determinants of health, lifestyle-related factors such as chronic stress, multimorbidities) in driving differences in patterns of cognitive decline between these subgroups.
PREDICTING RISK OF DISEASE FOR PRECISION MEDICINE
Our ultimate goal is to be able to devise scalable models that are tailored for a specific individual given their genetic profile, sex, lifestyle, previous medical history and environmental conditions. We have previously developed Bayesian generalized additive models to predict risk of breast cancer wherein we incorporated interactions between multiple omics such as gene expression and copy number variation. We seek to develop novel models that incorporate multiple layers of data to predict risk of cognitive decline.