{"id":{"repo_id":"odu","oai_identifier":"oai:digitalcommons.odu.edu:computerscience_etds-1198"},"canonical_url":"https://search.ndltd.org/etd/odu/oai:digitalcommons.odu.edu:computerscience_etds-1198","repository":{"repo_id":"odu","name":"Old Dominion University","base_url":"https://digitalcommons.odu.edu/do/oai/"},"display":{"title":"Learning Techniques in Prediction of Functional Epigenomic Events","abstract":"Accurately predicting functional epigenomic events from DNA sequences is critical to understanding gene regulation and the functional impact of non-coding variants. Despite considerable progress, critical challenges hamper the effectiveness and efficiency of existing deep learning approaches. These challenges include negative transfer in multi-task learning (MTL), suboptimal network architectures, and pervasive label noise, particularly the positive-unlabeled problem arising from data sparsity in single-cell assays. This dissertation presents a cohesive framework of novel learning techniques to effectively address these challenges. First, a highly scalable task grouping framework is presented to mitigate negative transfer in deep MTL. This method clusters tasks based on learned classification head weights, enabling effective in-group joint training across hundreds of epigenomic profiles. Second, this dissertation introduces the Efficient DNA Sequence Learner (EDSL), a biologicallyinformed neural architecture that incorporates variable-size convolutional filters and dense connections to enhance sequence pattern recognition and learning efficiency. Third, building on this foundation, a novel positive-anchor contrastive learning approach is presented, which is specifically tailored to the positive-unlabled setting of chromatin accessibility data. By adapting the supervised contrastive objective via a customized loss function that uses only labeled positives as anchors, this method leverages only reliable positive pairs, preventing model degradation from noisy negative labels. This contrastive loss is jointly trained with a class-balanced cross-entropy classifier to further refine the decision boundary. Empirical evaluations across large-scale datasets demonstrate that each contribution significantly improves predictive performance, robustness, and model interpretability, offering a powerful, integrated toolkit for functional genomics and variant prioritization.","abstract_html":"Accurately predicting functional epigenomic events from DNA sequences is critical to understanding gene regulation and the functional impact of non-coding variants. Despite considerable progress, critical challenges hamper the effectiveness and efficiency of existing deep learning approaches. These challenges include negative transfer in multi-task learning (MTL), suboptimal network architectures, and pervasive label noise, particularly the positive-unlabeled problem arising from data sparsity in single-cell assays. This dissertation presents a cohesive framework of novel learning techniques to effectively address these challenges. First, a highly scalable task grouping framework is presented to mitigate negative transfer in deep MTL. This method clusters tasks based on learned classification head weights, enabling effective in-group joint training across hundreds of epigenomic profiles. Second, this dissertation introduces the Efficient DNA Sequence Learner (EDSL), a biologicallyinformed neural architecture that incorporates variable-size convolutional filters and dense connections to enhance sequence pattern recognition and learning efficiency. Third, building on this foundation, a novel positive-anchor contrastive learning approach is presented, which is specifically tailored to the positive-unlabled setting of chromatin accessibility data. By adapting the supervised contrastive objective via a customized loss function that uses only labeled positives as anchors, this method leverages only reliable positive pairs, preventing model degradation from noisy negative labels. This contrastive loss is jointly trained with a class-balanced cross-entropy classifier to further refine the decision boundary. Empirical evaluations across large-scale datasets demonstrate that each contribution significantly improves predictive performance, robustness, and model interpretability, offering a powerful, integrated toolkit for functional genomics and variant prioritization.","abstract_has_math":false,"creators":["Shiri, Mohammad"],"institution":null,"degree_name":"Doctor of Philosophy (PhD)","degree_level":"Dissertation","degree_discipline":"Computer Science","degree_department":null,"school":null,"contributors":["Jiangwen Sun","Jing He","Lusi Li","Xing Fu"],"advisors":[],"committee_chairs":[],"committee_members":[],"year":null,"date_issued":"","date_published":null,"updated_at":"2026-11-19T08:34:20Z","subjects":["Multi-task learning","DNA sequences","Deep learning approaches","Robustness","Artificial Intelligence and Robotics","Computer Sciences"],"languages":[],"rights":["<p>In Copyright. URI: <a href=\"http://rightsstatements.org/vocab/InC/1.0/\" target=\"_blank\">http://rightsstatements.org/vocab/InC/1.0/</a> This Item is protected by copyright and/or related rights. You are free to use this Item in any way that is permitted by the copyright and related rights legislation that applies to your use. For other uses you need to obtain permission from the rights-holder(s).</p>"],"rights_urls":[],"identifier_entries":[{"key":"dc:identifier","label":"Identifier","values":["9798197810168"],"render_values":[{"text":"9798197810168","href":null,"code":true}]}],"person_identifier_entries":[]},"links":{"outbound_url":"https://digitalcommons.odu.edu/computerscience_etds/198","outbound_label":"Repository record","outbound_source":"dc:identifier","record_url":"https://digitalcommons.odu.edu/computerscience_etds/198","record_label":"Repository record","record_kind":"record","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://digitalcommons.odu.edu/do/oai/?verb=GetRecord&metadataPrefix=oai_etdms&identifier=oai%3Adigitalcommons.odu.edu%3Acomputerscience_etds-1198","prefix":"oai_etdms"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor","label":"Contributor","values":["Jiangwen Sun","Jing He","Lusi Li","Xing Fu"]},{"key":"dc:creator","label":"Author","values":["Shiri, Mohammad"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date.available","label":"Dc Date Available","values":["2026-07-15T07:00:00Z"]},{"key":"dc:date.created","label":"Dc Date Created","values":["2026-04-01T07:00:00Z"]},{"key":"thesis:degree_discipline","label":"Discipline","values":["Computer Science"]},{"key":"thesis:degree_level","label":"Degree Level","values":["Dissertation"]},{"key":"thesis:degree_name","label":"Degree Name","values":["Doctor of Philosophy (PhD)"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["Multi-task learning","DNA sequences","Deep learning approaches","Robustness","Artificial Intelligence and Robotics","Computer Sciences"]}]},{"id":"language_rights","label":"Language and Rights","entries":[{"key":"dc:rights","label":"Dc Rights","values":["<p>In Copyright. URI: <a href=\"http://rightsstatements.org/vocab/InC/1.0/\" target=\"_blank\">http://rightsstatements.org/vocab/InC/1.0/</a> This Item is protected by copyright and/or related rights. You are free to use this Item in any way that is permitted by the copyright and related rights legislation that applies to your use. For other uses you need to obtain permission from the rights-holder(s).</p>"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier","label":"Identifier","values":["9798197810168","https://digitalcommons.odu.edu/computerscience_etds/198"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description.abstract","label":"Abstract","values":["Accurately predicting functional epigenomic events from DNA sequences is critical to understanding gene regulation and the functional impact of non-coding variants. Despite considerable progress, critical challenges hamper the effectiveness and efficiency of existing deep learning approaches. These challenges include negative transfer in multi-task learning (MTL), suboptimal network architectures, and pervasive label noise, particularly the positive-unlabeled problem arising from data sparsity in single-cell assays. This dissertation presents a cohesive framework of novel learning techniques to effectively address these challenges. First, a highly scalable task grouping framework is presented to mitigate negative transfer in deep MTL. This method clusters tasks based on learned classification head weights, enabling effective in-group joint training across hundreds of epigenomic profiles. Second, this dissertation introduces the Efficient DNA Sequence Learner (EDSL), a biologicallyinformed neural architecture that incorporates variable-size convolutional filters and dense connections to enhance sequence pattern recognition and learning efficiency. Third, building on this foundation, a novel positive-anchor contrastive learning approach is presented, which is specifically tailored to the positive-unlabled setting of chromatin accessibility data. By adapting the supervised contrastive objective via a customized loss function that uses only labeled positives as anchors, this method leverages only reliable positive pairs, preventing model degradation from noisy negative labels. This contrastive loss is jointly trained with a class-balanced cross-entropy classifier to further refine the decision boundary. Empirical evaluations across large-scale datasets demonstrate that each contribution significantly improves predictive performance, robustness, and model interpretability, offering a powerful, integrated toolkit for functional genomics and variant prioritization."]},{"key":"dc:title","label":"Title","values":["Learning Techniques in Prediction of Functional Epigenomic Events"]}]}],"canonical_facts":{"dc:contributor":["Jiangwen Sun","Jing He","Lusi Li","Xing Fu"],"dc:creator":["Shiri, Mohammad"],"dc:date.available":["2026-07-15T07:00:00Z"],"dc:date.created":["2026-04-01T07:00:00Z"],"dc:description.abstract":["Accurately predicting functional epigenomic events from DNA sequences is critical to understanding gene regulation and the functional impact of non-coding variants. Despite considerable progress, critical challenges hamper the effectiveness and efficiency of existing deep learning approaches. These challenges include negative transfer in multi-task learning (MTL), suboptimal network architectures, and pervasive label noise, particularly the positive-unlabeled problem arising from data sparsity in single-cell assays. This dissertation presents a cohesive framework of novel learning techniques to effectively address these challenges. First, a highly scalable task grouping framework is presented to mitigate negative transfer in deep MTL. This method clusters tasks based on learned classification head weights, enabling effective in-group joint training across hundreds of epigenomic profiles. Second, this dissertation introduces the Efficient DNA Sequence Learner (EDSL), a biologicallyinformed neural architecture that incorporates variable-size convolutional filters and dense connections to enhance sequence pattern recognition and learning efficiency. Third, building on this foundation, a novel positive-anchor contrastive learning approach is presented, which is specifically tailored to the positive-unlabled setting of chromatin accessibility data. By adapting the supervised contrastive objective via a customized loss function that uses only labeled positives as anchors, this method leverages only reliable positive pairs, preventing model degradation from noisy negative labels. This contrastive loss is jointly trained with a class-balanced cross-entropy classifier to further refine the decision boundary. Empirical evaluations across large-scale datasets demonstrate that each contribution significantly improves predictive performance, robustness, and model interpretability, offering a powerful, integrated toolkit for functional genomics and variant prioritization."],"dc:identifier":["9798197810168","https://digitalcommons.odu.edu/computerscience_etds/198"],"dc:rights":["<p>In Copyright. URI: <a href=\"http://rightsstatements.org/vocab/InC/1.0/\" target=\"_blank\">http://rightsstatements.org/vocab/InC/1.0/</a> This Item is protected by copyright and/or related rights. You are free to use this Item in any way that is permitted by the copyright and related rights legislation that applies to your use. For other uses you need to obtain permission from the rights-holder(s).</p>"],"dc:subject":["Multi-task learning","DNA sequences","Deep learning approaches","Robustness","Artificial Intelligence and Robotics","Computer Sciences"],"dc:title":["Learning Techniques in Prediction of Functional Epigenomic Events"],"thesis:degree_discipline":["Computer Science"],"thesis:degree_level":["Dissertation"],"thesis:degree_name":["Doctor of Philosophy (PhD)"]},"updated_at":"2026-11-19T08:34:20Z"}