{"id":{"repo_id":"msu","oai_identifier":"oai:d.lib.msu.edu:etd_52860"},"canonical_url":"https://search.ndltd.org/etd/msu/oai:d.lib.msu.edu:etd_52860","repository":{"repo_id":"msu","name":"Michigan State University","base_url":"https://d.lib.msu.edu/oai"},"display":{"title":"Statistical and Deep Learning Methods to Decode Tissue Architecture in Spatial Transcriptomics : Cell Type Deconvolution and Spatial Domain Detection","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":["Xie, Aoqi"],"institution":null,"degree_name":null,"degree_level":null,"degree_discipline":null,"degree_department":null,"school":null,"contributors":["Cui, Yuehua","Weng, Haolei","Xie, Yuying","Chen, Bin"],"advisors":[],"committee_chairs":[],"committee_members":[],"year":2025,"date_issued":"2025","date_published":"2025","updated_at":"2026-11-18T18:59:31Z","subjects":["Statistics"],"languages":["eng"],"rights":["Attribution 4.0 International"],"rights_urls":[],"identifier_entries":[{"key":"dc:identifier","label":"Identifier","values":["etd:52860","local:Xie_grad.msu_0128D_21850"],"render_values":[{"text":"etd:52860","href":null,"code":true},{"text":"local:Xie_grad.msu_0128D_21850","href":null,"code":true}]}],"person_identifier_entries":[]},"links":{"outbound_url":"https://doi.org/doi:10.25335/d02z-0c76","outbound_label":"DOI","outbound_source":"dc:identifier","record_url":"https://doi.org/doi:10.25335/d02z-0c76","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_52860","prefix":"oai_dc"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor","label":"Contributor","values":["Cui, Yuehua","Weng, Haolei","Xie, Yuying","Chen, Bin"]},{"key":"dc:creator","label":"Author","values":["Xie, Aoqi"]}]},{"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:52860","local:Xie_grad.msu_0128D_21850","https://doi.org/doi:10.25335/d02z-0c76"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description","label":"Description","values":["Thesis (Ph.D.)--Michigan State University. Statistics - Doctor of Philosophy, 2025","This thesis aims to better understand how different cell types are arranged in tissue and how they form meaningful regions. Modern spatial gene expression technologies measure gene activity at many locations in a tissue, but each location often contains a mixture of cells. This makes it difficult to know which cell types are present and how they are organized. To address this, I developed two methods to estimate the mixture of cell types at each location, and one method to identify spatial regions that explicitly uses this cell type information.The first project introduces gwSPADE, a gene frequency-weighted reference-free SPAtial DEconvolution method for spatial transcriptomics data. It requires only the gene count matrix and utilizes appropriate weighting schemes within a topic model to accurately recover cell-type transcriptional profiles and their proportions at each spatial locationThrough simulations andmultiple real tissue datasets, gwSPADE shows higher accuracy than existing methods that also avoid external references. It recovers clearer cell type maps and reveals subtle differences within what was previously treated as a single cell type, improving our understanding of tissue structure and disease-related changes.The second project, gnSPADE, extends this idea by explicitly using known relationships between genes, such as genes that often work together in the same biological process. By encouraging related genes to contribute together to the same cell type, gnSPADE produces more coherent and stable cell type profiles and proportions. Across both simulated and real datasets from different technologies and tissues, gnSPADE improves accuracy and makes the inferred cell types easier to interpret biologically, especially when external reference data are missing, incomplete, or poorly matched to the tissue under study.While applying these two methods, we observed that the inferred cell types themselves form clear spatial patterns, and that these patterns strongly influence how experts define tissue regions such as layers, tumor cores, edges, and surrounding normal areas. Methods that ignore cell type composition often blur these boundaries. Motivated by this, the third project, CASTLE, focuses on spatial clustering. CASTLE builds a network that connects locations that are close in space and similar in composition or cell type, and then uses a graph neural network model to learn low-dimensional representations that are used to group locations into spatial domains. When explicit cell type labels are unavailable, CASTLE can fall back to grouping locations by overall gene expression similarity.Across many tissues and measurement platforms, CASTLE produces more accurate and stable spatial domains than existing approaches, recovers known anatomical layers, and reveals fine-grained structures and distinct tumor subregions linked to different gene expression programs. Together, these three projects provide a unified framework: gwSPADE and gnSPADE uncover what cell types and subtypes are present and where they are located, while CASTLE uses cell type information, along with spatial and expression patterns, to define meaningful tissue regions. Overall, these three projects offer complementary and interpretable tools for studying tissue organization, development, and disease, and point toward future directions such as incorporating image information, analyzing multiple samples together, and developing unified models that link cell type composition with spatial domains.","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 (147 p.) : ill."]},{"key":"dc:title","label":"Title","values":["Statistical and Deep Learning Methods to Decode Tissue Architecture in Spatial Transcriptomics : Cell Type Deconvolution and Spatial Domain Detection"]}]}],"canonical_facts":{"dc:contributor":["Cui, Yuehua","Weng, Haolei","Xie, Yuying","Chen, Bin"],"dc:creator":["Xie, Aoqi"],"dc:date":["2025"],"dc:description":["Thesis (Ph.D.)--Michigan State University. Statistics - Doctor of Philosophy, 2025","This thesis aims to better understand how different cell types are arranged in tissue and how they form meaningful regions. Modern spatial gene expression technologies measure gene activity at many locations in a tissue, but each location often contains a mixture of cells. This makes it difficult to know which cell types are present and how they are organized. To address this, I developed two methods to estimate the mixture of cell types at each location, and one method to identify spatial regions that explicitly uses this cell type information.The first project introduces gwSPADE, a gene frequency-weighted reference-free SPAtial DEconvolution method for spatial transcriptomics data. It requires only the gene count matrix and utilizes appropriate weighting schemes within a topic model to accurately recover cell-type transcriptional profiles and their proportions at each spatial locationThrough simulations andmultiple real tissue datasets, gwSPADE shows higher accuracy than existing methods that also avoid external references. It recovers clearer cell type maps and reveals subtle differences within what was previously treated as a single cell type, improving our understanding of tissue structure and disease-related changes.The second project, gnSPADE, extends this idea by explicitly using known relationships between genes, such as genes that often work together in the same biological process. By encouraging related genes to contribute together to the same cell type, gnSPADE produces more coherent and stable cell type profiles and proportions. Across both simulated and real datasets from different technologies and tissues, gnSPADE improves accuracy and makes the inferred cell types easier to interpret biologically, especially when external reference data are missing, incomplete, or poorly matched to the tissue under study.While applying these two methods, we observed that the inferred cell types themselves form clear spatial patterns, and that these patterns strongly influence how experts define tissue regions such as layers, tumor cores, edges, and surrounding normal areas. Methods that ignore cell type composition often blur these boundaries. Motivated by this, the third project, CASTLE, focuses on spatial clustering. CASTLE builds a network that connects locations that are close in space and similar in composition or cell type, and then uses a graph neural network model to learn low-dimensional representations that are used to group locations into spatial domains. When explicit cell type labels are unavailable, CASTLE can fall back to grouping locations by overall gene expression similarity.Across many tissues and measurement platforms, CASTLE produces more accurate and stable spatial domains than existing approaches, recovers known anatomical layers, and reveals fine-grained structures and distinct tumor subregions linked to different gene expression programs. Together, these three projects provide a unified framework: gwSPADE and gnSPADE uncover what cell types and subtypes are present and where they are located, while CASTLE uses cell type information, along with spatial and expression patterns, to define meaningful tissue regions. Overall, these three projects offer complementary and interpretable tools for studying tissue organization, development, and disease, and point toward future directions such as incorporating image information, analyzing multiple samples together, and developing unified models that link cell type composition with spatial domains.","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 (147 p.) : ill."],"dc:identifier":["etd:52860","local:Xie_grad.msu_0128D_21850","https://doi.org/doi:10.25335/d02z-0c76"],"dc:language":["eng"],"dc:relation":["Electronic Theses & Dissertations"],"dc:rights":["Attribution 4.0 International"],"dc:subject":["Statistics"],"dc:title":["Statistical and Deep Learning Methods to Decode Tissue Architecture in Spatial Transcriptomics : Cell Type Deconvolution and Spatial Domain Detection"],"dc:type":["text","Electronic dissertations"]},"updated_at":"2026-11-18T18:59:31Z"}