{"id":{"repo_id":"msu","oai_identifier":"oai:d.lib.msu.edu:etd_52697"},"canonical_url":"https://search.ndltd.org/etd/msu/oai:d.lib.msu.edu:etd_52697","repository":{"repo_id":"msu","name":"Michigan State University","base_url":"https://d.lib.msu.edu/oai"},"display":{"title":"BAYESIAN PARADIGMS IN STATISTICAL MODELING WITH APPLICATIONS TO GENOMICS DATA","abstract":"Thesis (Ph.D.)--Michigan State University. Statistics - Doctor of Philosophy, 2025","abstract_html":"Thesis (Ph.D.)--Michigan State University. Statistics - Doctor of Philosophy, 2025","abstract_has_math":false,"creators":["Kollipara, Hema Sri Sai"],"institution":null,"degree_name":null,"degree_level":null,"degree_discipline":null,"degree_department":null,"school":null,"contributors":["Maiti, Tapabrata","Sakhanenko, Lyudmila","Boddeti, Vishnu","de los Campos, Gustavo"],"advisors":[],"committee_chairs":[],"committee_members":[],"year":2025,"date_issued":"2025","date_published":"2025","updated_at":"2026-11-18T18:59:26Z","subjects":["Statistics"],"languages":["eng"],"rights":["Attribution 4.0 International"],"rights_urls":[],"identifier_entries":[{"key":"dc:identifier","label":"Identifier","values":["etd:52697","local:Kollipara_grad.msu_0128D_21667"],"render_values":[{"text":"etd:52697","href":null,"code":true},{"text":"local:Kollipara_grad.msu_0128D_21667","href":null,"code":true}]}],"person_identifier_entries":[]},"links":{"outbound_url":"https://doi.org/doi:10.25335/vtap-7946","outbound_label":"DOI","outbound_source":"dc:identifier","record_url":"https://doi.org/doi:10.25335/vtap-7946","record_label":"Repository record","record_kind":"doi","record_source":"dc:identifier","download_url":null,"download_label":null,"download_source":null,"access_status":null,"embargo_until":null,"access_note":null},"source_record":{"url":"https://d.lib.msu.edu/oai?verb=GetRecord&metadataPrefix=oai_dc&identifier=oai%3Ad.lib.msu.edu%3Aetd_52697","prefix":"oai_dc"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor","label":"Contributor","values":["Maiti, Tapabrata","Sakhanenko, Lyudmila","Boddeti, Vishnu","de los Campos, Gustavo"]},{"key":"dc:creator","label":"Author","values":["Kollipara, Hema Sri Sai"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date","label":"Dc Date","values":["2025"]},{"key":"dc:relation","label":"Dc Relation","values":["Electronic Theses & Dissertations"]},{"key":"dc:type","label":"Dc Type","values":["text","Electronic dissertations"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["Statistics"]}]},{"id":"language_rights","label":"Language and Rights","entries":[{"key":"dc:language","label":"Dc Language","values":["eng"]},{"key":"dc:rights","label":"Dc Rights","values":["Attribution 4.0 International"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier","label":"Identifier","values":["etd:52697","local:Kollipara_grad.msu_0128D_21667","https://doi.org/doi:10.25335/vtap-7946"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description","label":"Description","values":["Thesis (Ph.D.)--Michigan State University. Statistics - Doctor of Philosophy, 2025","The field of genomic data analysis has experienced significant growth through the integration of Bayesian statistical methodologies, which offer principled frameworks for inference under uncertainty, particularly in high-dimensional and sparse data regimes. Genomic studies often involve thousands of molecular features measured across a relatively small number of biological samples, giving rise to challenges such as overfitting, collinearity, and signal heterogeneity. Bayesian models are well-suited to address these issues by incorporating prior information, enforcing sparsity, and quantifying uncertainty in a coherent manner. This dissertation explores the development and evaluation of Bayesian methods tailored to three central tasks in genomic data analysis: variable selection, multivariate regression, and dimensionality reduction for count data.Feature selection plays a critical role in discovering biomarkers associated with phenotypic outcomes, and Bayesian global-local shrinkage priors provide an adaptive mechanism for discerning informative signals from noise. We examine the theoretical and empirical behavior of various shrinkage families and propose refinements that improve model robustness in the presence of correlated predictors and group structure among covariates. These developments are motivated by real-world applications in transcriptomic and proteomic data analysis, where biological interpretability is essential for downstream validation.In the context of multi-response regression, where multiple phenotypes or experimental conditions are modeled simultaneously, we develop a class of Bayesian models based on spike-and-slab priors that account for shared structure across outcomes. Such models enable joint inference on predictor-outcome associations and accommodate both row- and column-level sparsity. We assess these models under different prior specifications and evaluate their sensitivity to hyperparameters through comprehensive simulation studies. A third major contribution of this dissertation involves the extension of Bayesian principal component analysis (PCA) to discrete and overdispersed count data, which arise frequently in modern genomic technologies such as single-cell RNA sequencing (scRNA-seq). Classical PCA techniques rely on Gaussian assumptions and are therefore inadequate for modeling zero-inflated, non-negative data. We introduce a novel Bayesian PCA framework based on Poisson and negative binomial likelihoods, incorporating hierarchical priors for sparsity and adaptivity. This approach enables the extraction of low-dimensional representations while accounting for the stochastic nature of count-based measurements. We demonstrate the utility of our method on semi-synthetic and real single-cell datasets, highlighting improvements in clustering accuracy and biological signal preservation.Throughout the dissertation, we emphasize the importance of model validation, interpretability, and computational scalability. The proposed methods are benchmarked through simulation experiments that vary key data-generating conditions such as signal sparsity, overdispersion, and noise level. Where applicable, we also apply these models to public genomic datasets, including bulk and single-cell expression data, to illustrate their practical relevance and limitations. Taken together, this work provides a unified Bayesian perspective on three major analytical tasks in genomics and contributes flexible, generalizable methodologies for understanding complex biological systems through high-dimensional data.These contributions not only expand the statistical toolkit for genomic research but also serve as a foundation for broader methodological innovation in the analysis of structured, sparse, and heterogeneous data. By situating the work within a principled Bayesian framework and by systematically evaluating its empirical performance, this dissertation bridges theoretical rigor with practical applicability. The proposed models are readily extensible to other domains involving similar data structures, such as neuroscience, epidemiology, and ecology, thereby amplifying the broader impact of this research within and beyond the realm of statistical genomics.","Description based on online resource. Title from PDF t.p. (Michigan State University Fedora Repository, viewed ).","Includes bibliographical references."]},{"key":"dc:format","label":"Dc Format","values":["application/pdf","1 online resource (115 p.) : ill."]},{"key":"dc:title","label":"Title","values":["BAYESIAN PARADIGMS IN STATISTICAL MODELING WITH APPLICATIONS TO GENOMICS DATA"]}]}],"canonical_facts":{"dc:contributor":["Maiti, Tapabrata","Sakhanenko, Lyudmila","Boddeti, Vishnu","de los Campos, Gustavo"],"dc:creator":["Kollipara, Hema Sri Sai"],"dc:date":["2025"],"dc:description":["Thesis (Ph.D.)--Michigan State University. Statistics - Doctor of Philosophy, 2025","The field of genomic data analysis has experienced significant growth through the integration of Bayesian statistical methodologies, which offer principled frameworks for inference under uncertainty, particularly in high-dimensional and sparse data regimes. Genomic studies often involve thousands of molecular features measured across a relatively small number of biological samples, giving rise to challenges such as overfitting, collinearity, and signal heterogeneity. Bayesian models are well-suited to address these issues by incorporating prior information, enforcing sparsity, and quantifying uncertainty in a coherent manner. This dissertation explores the development and evaluation of Bayesian methods tailored to three central tasks in genomic data analysis: variable selection, multivariate regression, and dimensionality reduction for count data.Feature selection plays a critical role in discovering biomarkers associated with phenotypic outcomes, and Bayesian global-local shrinkage priors provide an adaptive mechanism for discerning informative signals from noise. We examine the theoretical and empirical behavior of various shrinkage families and propose refinements that improve model robustness in the presence of correlated predictors and group structure among covariates. These developments are motivated by real-world applications in transcriptomic and proteomic data analysis, where biological interpretability is essential for downstream validation.In the context of multi-response regression, where multiple phenotypes or experimental conditions are modeled simultaneously, we develop a class of Bayesian models based on spike-and-slab priors that account for shared structure across outcomes. Such models enable joint inference on predictor-outcome associations and accommodate both row- and column-level sparsity. We assess these models under different prior specifications and evaluate their sensitivity to hyperparameters through comprehensive simulation studies. A third major contribution of this dissertation involves the extension of Bayesian principal component analysis (PCA) to discrete and overdispersed count data, which arise frequently in modern genomic technologies such as single-cell RNA sequencing (scRNA-seq). Classical PCA techniques rely on Gaussian assumptions and are therefore inadequate for modeling zero-inflated, non-negative data. We introduce a novel Bayesian PCA framework based on Poisson and negative binomial likelihoods, incorporating hierarchical priors for sparsity and adaptivity. This approach enables the extraction of low-dimensional representations while accounting for the stochastic nature of count-based measurements. We demonstrate the utility of our method on semi-synthetic and real single-cell datasets, highlighting improvements in clustering accuracy and biological signal preservation.Throughout the dissertation, we emphasize the importance of model validation, interpretability, and computational scalability. The proposed methods are benchmarked through simulation experiments that vary key data-generating conditions such as signal sparsity, overdispersion, and noise level. Where applicable, we also apply these models to public genomic datasets, including bulk and single-cell expression data, to illustrate their practical relevance and limitations. Taken together, this work provides a unified Bayesian perspective on three major analytical tasks in genomics and contributes flexible, generalizable methodologies for understanding complex biological systems through high-dimensional data.These contributions not only expand the statistical toolkit for genomic research but also serve as a foundation for broader methodological innovation in the analysis of structured, sparse, and heterogeneous data. By situating the work within a principled Bayesian framework and by systematically evaluating its empirical performance, this dissertation bridges theoretical rigor with practical applicability. The proposed models are readily extensible to other domains involving similar data structures, such as neuroscience, epidemiology, and ecology, thereby amplifying the broader impact of this research within and beyond the realm of statistical genomics.","Description based on online resource. Title from PDF t.p. (Michigan State University Fedora Repository, viewed ).","Includes bibliographical references."],"dc:format":["application/pdf","1 online resource (115 p.) : ill."],"dc:identifier":["etd:52697","local:Kollipara_grad.msu_0128D_21667","https://doi.org/doi:10.25335/vtap-7946"],"dc:language":["eng"],"dc:relation":["Electronic Theses & Dissertations"],"dc:rights":["Attribution 4.0 International"],"dc:subject":["Statistics"],"dc:title":["BAYESIAN PARADIGMS IN STATISTICAL MODELING WITH APPLICATIONS TO GENOMICS DATA"],"dc:type":["text","Electronic dissertations"]},"updated_at":"2026-11-18T18:59:26Z"}