Michael P Epstein

Emory University School of Medicine
Country flagUnited States

Research Interests

ApplyKite TwinLearns as you apply

Make your next move.

Prepare your application

Review drafts before you send them.

Explore related searches

Articles (12)

Background Attention-deficit/hyperactivity disorder (ADHD) is a common heritable neurodevelopmental disorder, affecting ∼7 million children (11.4%) in the U.S. However, ADHD’s underlying genetic architecture remains largely unknown. Transcriptome-wide association studies (TWAS), which integrate expression quantitative trait loci (eQTL) and GWAS summary data, can identify differentially expressed risk genes underlying complex phenotypes. Here we conduct a TWAS of ADHD using expression data from multiple brain tissues to improve understanding of the complex genetic architecture underlying this psychopathology. Methods We applied the TWAS framework OTTERS to train multiple gene expression imputation models using cis-eQTL summary statistics from MetaBrain for three brain regions: cortex (n=2,683), basal ganglia (n=208), and cerebellum (n=492), and GWAS summary statistics from the most recent meta-analysis of ADHD (n=225,534; case fraction =0.17). We further conducted fine-mapping, colocalization analysis, and functional enrichment analysis. Results We identified 29 significant TWAS risk genes for ADHD (11 in cortex, 4 in basal ganglia, and 14 in cerebellum). Six genes appear novel for ADHD ( MPL, C1orf210, MDFIC, NKX2-2, FAM183A, HIGD1A) while four genes were previously implicated in autism spectrum disorder ( XRN2 , KIZ , NKX2-4 , NKX2-2 ). Pathway analysis indicated cortex and basal ganglia were enriched for neurodevelopmental pathways and regulation of cell development, and the protein-protein interaction network was statistically significant (p=1.12E-04). Conclusion This multi-tissue TWAS refines the genetic architecture of ADHD by identifying genes whose genetically regulated expression is associated with risk, including six candidates not previously linked to ADHD. Together, these findings provide novel insights for potential targets in translational research and drug discovery.

Year:

2026

Background: Germline genetics may influence tumor molecular characteristics and ultimately cancer survival. Studies of tumor characteristics, including our epithelial ovarian cancer (EOC) studies of Black women in the United States, may have RNA sequencing (RNA-seq) data from archival tumor tissue but lack germline DNA for at least some individuals. Incomplete germline DNA measurements impede analyses of important measures such as global genetic ancestry, often used in downstream analyses, by reducing sample sizes. Methods: The study population consists of 184 women who participated in two population-based studies of EOC with both germline and formalin-fixed, paraffin-embedded (FFPE) tumor samples and an additional 58 women diagnosed with EOC from the same two studies with only FFPE tumor tissue. We used tumor RNA-seq data to calculate proportions of African, European, and Asian genetic ancestry using a pipeline built on the packages SeqKit, HISAT2, SAMtools, BCFtools, PLINK, and ADMIXTURE. Women from the 1000 Genomes Project were used as the reference populations, and germline genetic ancestry estimates from blood or saliva were used as the baseline comparison. We evaluated multiple quality control strategies to improve genetic ancestry estimation. Results: Correlations between tumor RNA-seq–derived estimates of genetic ancestry from our pipeline and germline-derived African and European genetic ancestry ranged between 0.76 and 0.94. Conclusions: RNA-seq data from archival FFPE tumor tissue can be confidently and efficiently used to approximate global genetic ancestry in an admixed population when germline DNA is unavailable. Impact: This approach supports analyses of genetic ancestry and cancer when germline samples are not available.

Year:

2025

Supplementary Table S2 shows the distribution of SNP type stratified by whether or not data from a participant's tumor sample met the proposed SNP threshold to improve genetic ancestry estimation and overall.

Year:

2025

Supplementary Figure S1 shows the difference in distributions of African genetic ancestry subpopulation proportions as calculated from germline DNA and tumor RNASeq (n = 184; ESN = Esan, GWD = Gambian, LWK = Luhya, MSL = Mende, YRI = Yoruba).

Year:

2025

Differential co-expression analysis (DCA) aims to identify genes in a pathway whose shared expression depends on a risk factor. While DCA provides insights into the biological activity of diseases, existing methods are limited to categorical risk factors and/or suffer from bias due to batch and variance-specific effects. We propose a new framework, Kernel-based Differential Co-expression Analysis (KDCA), that harnesses correlation patterns between genes in a pathway to detect differential co-expression arising from general (i.e., continuous, discrete, or categorical) risk factors. Using various simulated pathway architectures, we find that KDCA accounts for common sources of bias to control the type I error rate while substantially increasing the power compared to the standard eigengene approach. We then applied KDCA to The Cancer Genome Atlas thyroid data set and found several differentially co-expressed pathways by age of diagnosis and BRAF mutation status that were undetected by the eigengene method. Collectively, our results demonstrate that KDCA is a powerful testing framework that expands DCA applications in expression studies.

Year:

2024

A transcriptome-wide association study (TWAS) is a popular statistical method for identifying genes whose genetically-regulated expression (GReX) component is associated with a trait of interest. Most TWAS approaches fundamentally assume that the training dataset (used to fit the gene expression prediction model) and testing GWAS dataset are from the same ancestrally homogenous population. If this assumption is violated, studies have shown a marked negative impact on expression prediction accuracy as well as reduced power of the downstream genetrait association test. These issues pose a particular problem for admixed individuals, whose genomes represent a mosaic of multiple continental ancestral segments. To resolve these issues, we present CADET, which enables powerful TWAS of admixed cohorts leveraging the local-ancestry (LA) information of the cohort along with summary-level eQTL data from reference panels of different ancestral groups. CADET combines multiple polygenic risk score models based on the summary-level eQTL reference data to predict LA-aware GReX components in admixed test samples. Using simulated data, we compare the imputation accuracy, power, and type I error rate of our proposed LA-aware approach to LA-unaware methods for performing TWASs. We show that CADET performs optimally in nearly all settings regardless of whether the genetic architecture of gene expression is dependent or independent of ancestry. We further illustrate CADET by performing a TWAS of 29 common blood biochemistry phenotypes within an admixed cohort from the UK Biobank and identify 18 hits unique to our LA-aware strategy with the majority of hits supported by existing GWAS findings.

Year:

2024

Family-based heritability estimates of complex traits are often considerably larger than their single-nucleotide polymorphism (SNP) heritability estimates. This discrepancy may be due to non-additive effects of genetic variation, including variation that interacts with other genes or environmental factors to influence the trait. Variance-based procedures provide a computationally efficient strategy to screen for SNPs with potential interaction effects without requiring the specification of the interacting variable. While valuable, such variance-based tests consider only a single trait and ignore likely pleiotropy among related traits that, if present, could improve power to detect such interaction effects. To fill this gap, we propose SCAMPI (Scalable Cauchy Aggregate test using Multiple Phenotypes to test Interactions), which screens for variants with interaction effects across multiple traits. SCAMPI is motivated by the observation that SNPs with pleiotropic interaction effects induce genotypic differences in the patterns of correlation among traits. By studying such patterns across genotype categories among multiple traits, we show that SCAMPI has improved performance over traditional univariate variance-based methods. Like those traditional variance-based tests, SCAMPI permits the screening of interaction effects without requiring the specification of the interaction variable and is further computationally scalable to biobank data. We employed SCAMPI to screen for interacting SNPs associated with four lipid-related traits in the UK Biobank and identified multiple gene regions missed by existing univariate variance-based tests. SCAMPI is implemented in software for public use.

Year:

2024

Multi-modal spatial omics data are invaluable for exploring complex cellular behaviors in diseases from both morphological and molecular perspectives. Current analytical methods primarily focus on clustering and classification, and do not adequately examine the relationship between cell morphology and molecular dynamics. Here, we present MorphLink, a framework designed to systematically identify disease-related morphological-molecular interplays. MorphLink has been evaluated across a wide array of datasets, showcasing its effectiveness in extracting and linking interpretable morphological features with various molecular measurements in multi-modal spatial omics analyses. These linkages provide a transparent depiction of cellular behaviors that drive transcriptomic heterogeneity and immune diversity across different regions within diseased tissues, such as cancer. Additionally, MorphLink is scalable and robust against cross-sample batch effects, making it an efficient method for integrative spatial omics data analysis across samples, cohorts, and modalities, and enhancing the interpretation of results for large-scale studies.

Year:

2024

Structural birth defects affect 3-4% of all live births and, depending on the type, tend to manifest in a sex-biased manner. Orofacial clefts (OFCs) are the most common craniofacial structural birth defects and are often divided into cleft lip with or without cleft palate (CL/P) and cleft palate only (CP). Previous studies have found sex-specific risks for CL/P, but these risks have yet to be evaluated in CP. CL/P is more common in males and CP is more frequently observed in females, so we hypothesized there would also be sex-specific differences for CP. Using a trio-based cohort, we performed sex-stratified genome-wide association studies (GWAS) based on proband sex followed by a genome-wide gene-by-sex (GxS) interaction testing. There were 13 loci significant for GxS interactions, with the top finding in LTBP1 (RR=3.37 [2.04 - 5.56], p=1.93x10 -6 ). LTBP1 plays a role in regulating TGF-B bioavailability, and knockdown in both mice and zebrafish lead to craniofacial anomalies. Further, there is evidence for differential expression of LTBP1 between males and females in both mice and humans. Therefore, we tested the association between the imputed genetically regulated gene expression of genes with significant GxS interactions and the CP phenotype. We found significant association for LTBP1 in cell cultured fibroblasts in female probands (p=0.0013) but not in males. Taken altogether, we show there are sex-specific risks for CP that are otherwise undetectable in a combined sex cohort, and LTBP1 is a candidate risk gene, particularly in females.

Year:

2024

The "missing" heritability of complex traits may be partly explained by genetic variants interacting with other genes or environments that are difficult to specify, observe, and detect. We propose a new kernel-based method called Latent Interaction Testing (LIT) to screen for genetic interactions that leverages pleiotropy from multiple related traits without requiring the interacting variable to be specified or observed. Using simulated data, we demonstrate that LIT increases power to detect latent genetic interactions compared to univariate methods. We then apply LIT to obesity-related traits in the UK Biobank and detect variants with interactive effects near known obesity-related genes (URL: https://CRAN.R-project.org/package=lit ).

Year:

2024

Year:

2024

Year:

2023

Collaborators (9)

Aliza Wingo

Albert Holmes Rowe Endowed Chair of Genetics I

University of California, Davis

UNITED STATES
View details

Jingjing Yang

Associate Professor

Emory University School of Medicine

UNITED STATES
View details

Joellen Schilkdraut

Professor

Emory University

UNITED STATES
View details

David Cutler

Professor

Emory University School of Medicine

UNITED STATES
View details

Casey Greene

Director

University of Colorado Anschutz Medical Campus

UNITED STATES
View details

Jennifer Anne Doherty

Professor

University of Utah

UNITED STATES
View details

Taylor Head

The University of Texas MD Anderson Cancer Center

UNITED STATES
View details

Thomas Wingo

University of California, Davis

UNITED STATES
View details

Natalie Davidson

Assistant Professor

University of Colorado Anschutz Medical Campus

UNITED STATES
View details