About

Professor Sunduz Keles is a core faculty member at the Center for Genomic Science Innovation at the University of Wisconsin–Madison. Her academic credentials are in Statistics, Biostatistics & Medical Informatics. Her research focuses on developing statistical genomic methods, with a particular emphasis on noncoding regulatory sequences and the effects of their variation. This work contributes to advancing the understanding of genome regulation and the impact of genetic variation beyond coding regions, which is critical for interpreting complex genomic data and its implications in biology and medicine.

Research topics

• Biology
• Genetics
• Cell biology
• Computational biology
• Cancer research
• Evolutionary biology
• Endocrinology
• Immunology

Selected publications

• Systematic background selection with BasCoD enhances contrastive dimension reduction in single cell genomics

  Nature Communications · 2026-03-17

  articleOpen accessSenior authorCorresponding

  In single-cell experiments spanning diverse conditions, distinguishing variation specific to one condition (e.g., treatment) from shared or background variation (e.g., control) is critical for uncovering treatment-specific molecular responses. However, these studies typically yield ultra-high-dimensional data, necessitating effective dimension reduction for reliable biological interpretation. Contrastive dimension reduction methods address this challenge by identifying low-dimensional features enriched in a target dataset relative to a background dataset that captures shared variation. Despite their growing utility, the success of such methods critically depends on the choice of background, yet no formal criterion exists for evaluating or selecting backgrounds. To address this gap, we introduce BasCoD, a statistical testing framework based on spectral subspace inclusion theory, that enables rigorous evaluation and systematic selection of background datasets. Applying BasCoD across a range of single-cell datasets, we show that it effectively identifies suitable backgrounds, substantially improving the contrast and interpretability of the resulting target representations. We further demonstrate how BasCoD can guide the design of contrastive analyses in large-scale single-cell experiments conducted under heterogeneous conditions and elucidate potential interaction effects in perturbation studies. The efficiency of contrastive dimension reduction depends on selecting a valid background for a target dataset. Here, the authors introduce BasCoD, a statistical testing framework that evaluates background validity and improves contrastive dimension reduction in single-cell genomics applications.

  PublisherDOI

• Causal gene regulatory network inference from Perturb-seq via adaptive instrumental variable modeling

  bioRxiv (Cold Spring Harbor Laboratory) · 2026-02-19

  articleOpen accessSenior authorCorresponding

  Abstract Inferring causal gene regulatory networks (GRNs) from observational single-cell data is challenging due to confounding. While Perturb-seq provides causal leverage, existing methods are often biased by heterogeneous CRISPRi knockdown efficiencies and restrictive assumptions like acyclicity. We present ADAPRE, a framework that treats CRISPR interventions as instrumental variables within a Poisson-lognormal model. By adaptively accounting for variable perturbation strength, ADAPRE recovers potentially cyclic structures and outperforms existing methods. Applied to a genome-wide K562 Perturb-seq dataset, it reconstructs networks enriched for known biological interactions and identifies coherent, leukemia-associated subnetworks, establishing a scalable approach for causal GRN inference.

  PublisherDOI

• CMAPS: Causal Mediation Analysis of Perturbation Screens with Application to Genome-scale Perturb-seq Data

  bioRxiv (Cold Spring Harbor Laboratory) · 2026-05-23

  articleOpen accessSenior authorCorresponding

  CRISPR-Cas9 perturbation screens coupled with single-cell multi-omic profiling enable dissection of gene regulatory mechanisms, yet existing analyses largely quantify total perturbation effects and offer limited insight into the molecular intermediates that transmit these effects. We introduce CMAPS (Causal Mediation Analysis for Perturbation Screens), a semiparametric framework for robust mediation analysis that accommodates unmeasured mediator-outcome confounding and incorporates an adaptive bootstrap test with false discovery rate control. Simulations and data-driven computational experiments show that CMAPS yields accurate, calibrated mediation estimates and robust mediator identification, as confirmed through negative controls and permutation-based validation. Applied to K562 Perturb-seq, CMAPS recapitulates transcriptional cascades downstream of GATA1. In BT16 MultiPerturb-seq data, CMAPS identifies promoter-centric, enhancer-distributed, and mixed cis -regulatory programs linking chromatin remodeling factors to transcriptional responses. CMAPS provides a rigorous and interpretable framework for mechanistic inference in single-cell perturbation screens. CMAPS is implemented in R and is available at https://github.com/keleslab/CMAPS.

  PublisherOA PDFDOI

• Systematic Background Selection for Enhanced Contrastive Dimension Reduction

  bioRxiv (Cold Spring Harbor Laboratory) · 2025-05-15

  preprintOpen accessSenior authorCorresponding

  Contrastive dimension reduction enhances the analysis of high-dimensional data by generating target-specific low-dimensional representations relative to a background. Emerging methods for contrastive dimension reduction have demonstrated their utility in extracting target-specific signals across domains involving high-dimensional observations, including genomics, transcriptomics, and pattern recognition. However, even though choosing an appropriate background is critical to the success of contrastive dimension reduction, no established criterion currently exists for selecting such backgrounds. To address this gap, we introduce BasCoD, a novel testing framework grounded in the spectral theory of subspace inclusion, to enable rigorous evaluation and optimal selection of backgrounds. Extensive application of BasCoD across diverse single-cell datasets demonstrates its effectiveness in systematically identifying suitable backgrounds, thereby significantly improving the contrast and interpretability of the derived target representations. We further illustrate how BasCoD can facilitate design of appropriate backgrounds in large-scale single cell experiments under heterogeneous conditions.

  PublisherOA PDFDOI

• Whole genome methylation sequencing in blood from persons with mild cognitive impairment and dementia due to Alzheimer's disease identifies cognitive status

  Alzheimer s & Dementia · 2025-01-01 · 7 citations

  articleOpen access

  INTRODUCTION: Whole genome methylation sequencing (WGMS) in blood identifies differential DNA methylation in persons with late-onset dementia due to Alzheimer's disease (AD) but has not been tested in persons with mild cognitive impairment (MCI). METHODS: We used WGMS to compare DNA methylation levels at 25,244,219 CpG loci in 382 blood samples from 99 persons with MCI, 109 with AD, and 174 who are cognitively unimpaired (CU). RESULTS: WGMS identified 9756 differentially methylated positions (DMPs) in persons with MCI, including 1743 differentially methylated genes encoding proteins in biological pathways related to synapse organization, dendrite development, and ion transport. A total of 447 DMPs exhibit progressively increasing or decreasing DNA methylation levels among CU, MCI, and AD that correspond to cognitive status. DISCUSSION: WGMS identifies DMPs in known and newly detected genes in blood from persons with MCI and AD that support blood DNA methylation levels can distinguish cognitive status. HIGHLIGHTS: Whole genome methylation levels in blood from 99 persons with mild cognitive impairment (MCI), 109 with Alzheimer's disease, and 174 who are cognitively unimpaired were analyzed. Nine thousand seven hundred fifty-six differentially methylated positions (DMPs) were identified in MCI. One thousand seven hundred forty-three genes comprise one or more DMPs in persons with MCI. Fifty-eight DMPs and 392 differentially methylated genes are shared among the three pairwise comparisons. Four hundred forty-seven DMPs exhibit progressive changes that correspond to cognitive status.

  PublisherOA PDFDOI

• Inflammatory signaling selectively disrupts GATA factor-regulated networks governing erythroid differentiation and hemoglobin synthesis

  Blood · 2025-11-03

  articleOpen access

  Abstract Chronic inflammation, often associated with aging, instigates anemia of inflammation (AI) that afflicts hundreds of thousands in the USA (Ganz, N Engl J Med. 2019). Human erythroid cells express cytokine/chemokine receptor genes, yet erythroid-intrinsic actions of many receptor agonists are unstudied, even though they correlate with AI. Using primary human hematopoietic stem and progenitor cells (HSPCs) cultured ex vivo and flow cytometric assays, we discovered an amalgam (TNFα, IFNγ, and IL-6) of inflammatory mediators implicated in AI synergistically abrogated erythropoiesis. To dissect the mechanism, we treated human HSPCs with or without the amalgam and harvested cells for single-cell RNA-sequencing (scRNA-seq) on days 9 and 13. Cell types were annotated using Azimuth human bone marrow as a reference (Oetjen et al., JCI Insight. 2018; Granja et al. Nat Biotechnol. 2019). There were no major cellularity changes between vehicle- and inflammation-treated day 9 cells. On day 13, >80% of vehicle-treated cells were late erythroid with few non-erythroid cells. The majority of late erythroid cells were lost, whereas early erythroid and select myeloid cells were retained in inflammation-treated cells. These results, which were confirmed by flow cytometric analysis and Giemsa staining, suggest that erythroblasts are hypersensitive to inflammation. We performed differential gene expression (DEG) analysis in early and late erythroid cells using pseudobulk limma-voom workflow. Inflammation highly induced TNFα, IFNγ, and IL-6 targets established in non-erythroid systems in early and late populations. We asked if inflammation impacts transcription factors and cytokine receptors vital for erythroid survival and differentiation. Inflammation induced GATA2 in early (logFC=1.26, p-adj=5.17e-4) and late erythroid cells (logFC=0.94, p-adj=4.68e-3) and slightly reduced KLF1, LDB1, and EPOR; GATA1, ZFPM1, LMO2, NFE2, TAL1, and KIT were unaffected. Pseudotime analysis revealed GATA2 decreased over pseudotime and increased with inflammation. KLF1, LDB1 and EPOR decreased with inflammation, and GATA1 and KIT were constant. Inflammation increased GATA2 protein 2.0-fold (p=0.0001) without affecting GATA1. TNFα, but not IFNγ or IL-6, increased GATA2 (1.9-fold, p=0.0033), while GATA1 was unaltered. To test the hypothesis that GATA and inflammatory mechanisms intersect, we quantified expression of established targets. For the top 100 GATA1-regulated genes from G1E-ER-GATA1 rescue (Tanimura et al., EMBO Rep. 2016), inflammation mimicked GATA1 activation of select targets (e.g., ALAS2, SLC4A1, and HBE1) and antagonized GATA1 function to regulate another cohort, e.g., HBB and VIM implicated in terminal differentiation. GATA1 regulates 145 SLC genes in G1E-ER-GATA1 cells that mediate small molecule transport (Zwifelhofer et al., PLoS Genet. 2020). Inflammation promoted or antagonized GATA1 regulation of SLCs, including Zn2+ transporters SLC30A1 and SLC39A8 and adenosine transporters SLC29A1 and A2 with roles in erythroid survival and differentiation. Inflammation upregulated most GATA2-induced genes (GATA2 rescue in GATA2+/- HUDEP2), consistent with elevated GATA2 mRNA and protein. As inflammation antagonized GATA2 regulation of SLC4A1, PRG2, and CD69, inflammation promoted or inhibited GATA factor function in a context-dependent mechanism. DEG and pseudotime analysis revealed decreased HBB and increased HBG1 and HBG2 upon inflammation in late erythroid cells. Interrogation of established γ-globin regulators revealed BCL11A downregulation by inflammation in early (logFC=-1.17, p-adj=3.45e-3) and late erythroid cells (logFC=-0.92, p-adj=2.53e-3). Amalgam or IFNγ decreased BCL11A protein and increased γ-globin. BCL11A expression is activated by intronic enhancers that are gene editing targets for sickle cell disease and β-thalassemia (Canver et al., Nature. 2015). ATAC-seq revealed inflammation abrogated accessibility at all enhancers. Using gene editing strategies and multiomics, we are dissecting the underlying mechanisms. In summary, our multiomic analyses with normal human erythroid progenitor cells unveiled mechanistic intersections between GATA factor and inflammation networks governing erythropoiesis. The study provide mechanistic insights into the control of erythropoiesis under normal and pathological conditions and translational opportunities vis-à-vis anemia of inflammation.

  PublisherOA PDFDOI

• Sex-Specific Differential DNA Methylation in Mild Cognitive Impairment and Alzheimer’s Disease

  bioRxiv (Cold Spring Harbor Laboratory) · 2025-05-19 · 1 citations

  preprintOpen access

  Abstract Sex differences in late-onset Alzheimer’s disease (AD) progression include accelerated decrements in cognitive status and greater amyloid and tau biomarker burdens in females. To identify sex-specific differentially methylated positions (DMPs) and genes in persons with mild cognitive impairment (MCI) and AD, we analyzed whole genome methylation sequencing on blood samples from participants with MCI (N=99, 52% female), AD (N=109, 43% female), and those cognitively unimpaired (CU; N=174, 52% female). Ninety-four percent of DMPs from MCI vs . CU, AD vs . CU, and AD vs . MCI pairwise comparisons were sex-specific. Female-specific DMPs were enriched in neurologic gene sets ( e.g ., synaptic membrane, ion channel complex), while male-specific DMPs showed limited enrichment. Sex-specific DMPs overlapped blood-specific enhancers, promoters, and transcription factor binding motifs, highlighting divergent epigenetic regulation by sex. These findings identify sex-specific genes and molecular pathways in MCI and AD and support that blood DNA methylation levels can distinguish cognitive status.

  PublisherOA PDFDOI

• Oncogenic DEAD-box ATPase DDX41 establishes transcript ensembles via CLK3-dependent and -independent mechanisms

  Nature Communications · 2025-11-05 · 1 citations

  articleOpen access

  Post-transcriptional diversification of RNA transcripts mediated by complex processing machinery, including DEAD-box ATPases, establishes and maintains cellular phenotypes. For example, DDX41 controls RNA splicing, innate immune signaling, and genome stability. Although heterozygous DDX41 germline genetic variation occurs in familial myelodysplastic syndrome (MDS) and acute myeloid leukemia (AML), the DDX41 contributions to splicing globally, biological processes, and pathogenic mechanisms are incompletely defined. Using a genetic rescue system with Ddx41+/− myeloid progenitors, we established global wildtype DDX41 and pathogenic variant mechanisms. Differing from pathogenic variants of other RNA splicing regulators, DDX41 deficiency compromised multiple splicing steps. DDX41-regulated transcripts encoded factors controlling RNA splicing, including Cdc2-like kinase 3 (CLK3). DDX41 regulated Clk3 transcripts, and elevated CLK3 during myeloid differentiation. Loss-of-function analysis revealed DDX41-regulated splicing commonly, but not always, required CLK3. Thus, through a mechanism utilizing a splicing factor kinase that itself is DDX41-regulated, DDX41 establishes transcript ensembles in myeloid progenitors. Although DDX41 genetic variation occurs in myelodysplastic syndrome and acute myeloid leukemia, the pathogenic mechanisms remained unclear. Here, the authors found DDX41 regulates CLK3-dependent alternative splicing to establish transcript ensembles, while pathogenic variants limit the activity.

  PublisherOA PDFDOI

• FlyVISTA, an integrated machine learning platform for deep phenotyping of sleep in <i>Drosophila</i>

  Science Advances · 2025-03-12 · 20 citations

  articleOpen access

  There is great interest in using genetically tractable organisms such as Drosophila to gain insights into the regulation and function of sleep. However, sleep phenotyping in Drosophila has largely relied on simple measures of locomotor inactivity. Here, we present FlyVISTA, a machine learning platform to perform deep phenotyping of sleep in flies. This platform comprises a high-resolution closed-loop video imaging system, coupled with a deep learning network to annotate 35 body parts, and a computational pipeline to extract behaviors from high-dimensional data. FlyVISTA reveals the distinct spatiotemporal dynamics of sleep and wake-associated microbehaviors at baseline, following administration of the sleep-inducing drug gaboxadol, and with dorsal fan-shaped body drivers. We identify a microbehavior (“haltere switch”) exclusively seen during quiescence that indicates a deeper sleep stage. These results enable the rigorous analysis of sleep in Drosophila and set the stage for computational analyses of microbehaviors in quiescent animals.

  PublisherDOI

• Generation of Hoxa11-3XFLAG and Hoxd11-3XFLAG alleles to investigate Hox11 genome-wide binding

  Developmental Biology · 2025-05-17 · 1 citations

  article
  PublisherDOI

Recent grants

• NIH Grant R21HG009744

  NIH · $395k · 2020

• NIH Grant R21HG006716

  NIH · $368k · 2015

• Regularized Dimension Reduction for High Dimensional Data

  NSF · $100k · 2008–2011

• High dimensional statistical data modeling and integration for studying regulatory variation

  NIH · $4.8M · 2007–2026

• NIH Grant U01HG007019

  NIH · $1.5M · 2016

Frequent coauthors

• Emery H. Bresnick

  University of Wisconsin Carbone Cancer Center

  78 shared

• Kirby D. Johnson

  University of Wisconsin Carbone Cancer Center

  44 shared

• Ye Zheng

  Fred Hutch Cancer Center

  30 shared

• Öznur Taştan

  Sabancı Üniversitesi

  29 shared

• Burçak Otlu

  Middle East Technical University

  26 shared

• James B. Meigs

  25 shared

• Jane E. Freedman

  Vanderbilt University Medical Center

  25 shared

• Alexandra A. Soukup

  23 shared

Education

• Ph.D., Biostatistics

  University of Wisconsin–Madison

  2004

• M.S., Biostatistics

  University of Wisconsin–Madison

  2001

• B.S., Mathematics

  University of Wisconsin–Madison

  1998