Joshua Denny, M.D., M.S.

Dr. Denny’s laboratory seeks to discover gene-disease relationships by gathering, assessing, and analyzing the human phenome across genomics and environmental exposures. With this mission in mind, those in the Precision Health Informatics Section primarily repurpose health records as a source of longitudinal phenotype information and use links to genomic and other data. We partner with researchers to create trans-initiative resources cataloguing genetic associations across phenotypes.

The Precision Health Informatics Section relies on large scale data and bioinformatics approaches to more effectively characterize and identify genetic diseases. Electronic Health Records (EHRs), which typically include hospital billing codes, laboratory and vital signs, provider documentation, reports and tests, and medication records, are the major source of data for the laboratory. Over the years, studies have demonstrated that genetic analyses evaluating EHRs typically have larger sample sizes, are more cost-effective, and provide more opportunities for broad-ranging longitudinal investigations. In light of these facts, EHRs have become a powerful resource to illuminate shared genetic architecture across diseases through the development of genome-wide association studies (GWAS), phenome-wide association studies (PheWAS), transcriptome wide association studies (TWAS) and colocalization analyses, pharmacogenomic investigations, polygenic risk scores (PRS), phenotype risk scores (PheRS), Mendelian randomization (MR), biogeographic ancestry modeling, and exploration of the impact of rare disease variants. GWAS evaluates the association of millions of genetic variants with a particular disease while PheWAS examines the range of diseases associated with a particular genetic variant (or other analyte) to identify potentially pleiotropic relationships. These types of approaches, combined with biomedical and functional genomic informatics resources as well as innovative statistical modeling techniques, can elucidate genomic architecture of disease, common biological mechanisms underlying disease development and progression, and clinically relevant therapeutic targets.

Currently, the laboratory harnesses data from large scale biorepositories such as the eMERGE (Electronic Medical Records and Genetics) Network, BioVU, UK Biobank, Million Veteran Program (MVP), and All of Us . A common project for those in Dr. Denny’s group is to analyze complex datasets incorporating hundreds of thousands of predictors and up to millions of subjects. Their work accounts for complex interactions between a highly dimensional feature space through statistical and artificial intelligence (AI)/machine learning algorithms designed to process this complexity.

Additionally, members of the laboratory have incorporated dense, temporal data in intensive care environments and sparse, sporadically collected outpatient data on diverse and heterogeneous study populations. The nature of this research is based on highly imbalanced data with rare outcomes and events. In conjunction with these data sources and growing research partnerships, the laboratory aims to identify features that track with behavioral health traits (e.g., activity, sleep, imaging, etc.) and build novel phenotypes (e.g., predicted suicide risk, predicted carrier of genetic variant, etc.). EHRs offer a unique chance to evaluate a multitude of health outcomes including complex human disease, response to medication, clinical characteristics, and environmental influences impacting patient health for association with genetic factors.

While there is an overwhelming amount of information available in EHRs, typically this data is unstructured and common difficulties arise associated with data availability, missingness, and inconsistency. Therefore, another meaningful component of this laboratory’s mission is to extract practical information from the EHRs in a systematic and unbiased fashion. Dr. Denny and others have developed tools which facilitate data extraction, such as KnowledgeMap for natural language processing of clinical text and “phecodes” for phenotypic restructuring and harmonization from EHR billing codes. Moreover, Dr. Denny and others have leveraged data found in EHRs to facilitate the recognition of rare-variant associations. The aggregated risk scores, known as phenotype risk scores (PheRS), leverage structured EHR data by mapping known clinical characteristics of a given Mendelian disease (typically extracted as human phenotype ontology [HPO] terms associated with characteristics from Online Mendelian Inheritance in Man (OMIM)). These HPO terms are then mapped to phecodes (which are available for each person contained in an EHR) and aggregated into a risk score for each individual. PheRS has demonstrated effective identification of potentially pathogenic variants and has replicated previously unrecognized associations.

In summary, the Precision Health Informatics Lab proposes efficient and cost-effective approaches to identify novel disease/trait-variant relationships with pleiotropic impacts by evaluating large-scale EHR data from multiple sources. Their work has the potential to fundamentally augment the knowledgebase of human health through illumination of the genomic architecture of complex human diseases and traits, insight into common biological mechanisms underlying disease development and progression and trait distributions, and identification of clinically relevant therapeutic targets.