서브메뉴
검색
Dissecting the Function of the Non-Coding Genome Using Observational Data and Genomic Deep Learning Models
Dissecting the Function of the Non-Coding Genome Using Observational Data and Genomic Deep Learning Models
상세정보
- 자료유형
- 학위논문 서양
- 최종처리일시
- 20260202104825
- ISBN
- 9798293893133
- DDC
- 574
- 저자명
- Kathail, Pooja.
- 서명/저자
- Dissecting the Function of the Non-Coding Genome Using Observational Data and Genomic Deep Learning Models
- 발행사항
- [Sl] : University of California, Berkeley, 2025
- 발행사항
- Ann Arbor : ProQuest Dissertations & Theses, 2025
- 형태사항
- 111 p
- 주기사항
- Source: Dissertations Abstracts International, Volume: 87-04, Section: B.
- 주기사항
- Advisor: Ioannidis, Nilah;Ye, Jimmie.
- 학위논문주기
- Thesis (Ph.D.)--University of California, Berkeley, 2025.
- 초록/해제
- 요약Understanding the causes of disease is a critical step to improving human health. This thesis shows how large-scale genomic datasets and machine learning methods can be used to understand the role of non-protein-coding genetic mutations in disease. In Chapter 2, we collect and generate a population-scale single cell multiome dataset from a diverse human cohort, and use this data to identify non-coding mutations associated with differences in gene expression (eQTLs) and chromatin accessibility (caQTLs). We find that caQTLs often explain more autoimmune disease signal than eQTLs, due to their ability to tag primed chromatin states. Next, we evaluate a recent machine learning paradigm---genomic deep learning---that models the relationship between non-coding sequences and cell type specific molecular phenotypes, such as gene expression and chromatin accessibility. A promising application of such models is in silico prediction of non-coding mutation effects. In Chapter 3, we find that current genomic deep learning models perform poorly in cell type specific regulatory elements, which harbor a large fraction of the heritability of complex diseases. We identify model training strategies---such as single-task learning---to maximize performance in cell type specific regulatory elements. Finally, in Chapter 4, we find that current genomic deep learning models have high uncertainty in their predictions for out of distribution sequences containing genetic variants, suggesting that variant-based training data may be a path towards improved variant effect prediction. Together, this work analyzes novel data measuring the effect of non-coding mutations on cell type specific gene regulation, providing insight into the data modalities most informative for disease, and exposes important limitations of current machine learning tools, furthering our ability to understand the role of non-coding mutations in disease.
- 일반주제명
- Bioinformatics
- 일반주제명
- Cellular biology
- 일반주제명
- Genetics
- 키워드
- Genetic variants
- 키워드
- Machine learning
- 키워드
- Gene regulation
- 기타저자
- University of California, Berkeley Bioinformatics & Computational Biology
- 기본자료저록
- Dissertations Abstracts International. 87-04B.
- 전자적 위치 및 접속
- 로그인 후 원문을 볼 수 있습니다.
MARC
008260126s2025 us c eng d■001000017359041
■00520260202104825
■006m o d
■007cr#unu||||||||
■020 ▼a9798293893133
■035 ▼a(MiAaPQ)AAI32169864
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a574
■1001 ▼aKathail, Pooja.
■24510▼aDissecting the Function of the Non-Coding Genome Using Observational Data and Genomic Deep Learning Models
■260 ▼a[Sl]▼bUniversity of California, Berkeley▼c2025
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2025
■300 ▼a111 p
■500 ▼aSource: Dissertations Abstracts International, Volume: 87-04, Section: B.
■500 ▼aAdvisor: Ioannidis, Nilah;Ye, Jimmie.
■5021 ▼aThesis (Ph.D.)--University of California, Berkeley, 2025.
■520 ▼aUnderstanding the causes of disease is a critical step to improving human health. This thesis shows how large-scale genomic datasets and machine learning methods can be used to understand the role of non-protein-coding genetic mutations in disease. In Chapter 2, we collect and generate a population-scale single cell multiome dataset from a diverse human cohort, and use this data to identify non-coding mutations associated with differences in gene expression (eQTLs) and chromatin accessibility (caQTLs). We find that caQTLs often explain more autoimmune disease signal than eQTLs, due to their ability to tag primed chromatin states. Next, we evaluate a recent machine learning paradigm---genomic deep learning---that models the relationship between non-coding sequences and cell type specific molecular phenotypes, such as gene expression and chromatin accessibility. A promising application of such models is in silico prediction of non-coding mutation effects. In Chapter 3, we find that current genomic deep learning models perform poorly in cell type specific regulatory elements, which harbor a large fraction of the heritability of complex diseases. We identify model training strategies---such as single-task learning---to maximize performance in cell type specific regulatory elements. Finally, in Chapter 4, we find that current genomic deep learning models have high uncertainty in their predictions for out of distribution sequences containing genetic variants, suggesting that variant-based training data may be a path towards improved variant effect prediction. Together, this work analyzes novel data measuring the effect of non-coding mutations on cell type specific gene regulation, providing insight into the data modalities most informative for disease, and exposes important limitations of current machine learning tools, furthering our ability to understand the role of non-coding mutations in disease.
■590 ▼aSchool code: 0028.
■650 4▼aBioinformatics
■650 4▼aCellular biology
■650 4▼aGenetics
■653 ▼aChromatin accessibility
■653 ▼aGenetic mutations
■653 ▼aGenetic variants
■653 ▼aMachine learning
■653 ▼aGene regulation
■690 ▼a0715
■690 ▼a0379
■690 ▼a0369
■71020▼aUniversity of California, Berkeley▼bBioinformatics & Computational Biology.
■7730 ▼tDissertations Abstracts International▼g87-04B.
■790 ▼a0028
■791 ▼aPh.D.
■792 ▼a2025
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17359041▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.
