서브메뉴
검색
Feature-Based Deep Learning Approaches to Facilitate Drug Discovery
Feature-Based Deep Learning Approaches to Facilitate Drug Discovery
상세정보
- 자료유형
- 학위논문 서양
- 최종처리일시
- 20260202104701
- ISBN
- 9798280771284
- DDC
- 540
- 저자명
- Xue, Mingyi.
- 서명/저자
- Feature-Based Deep Learning Approaches to Facilitate Drug Discovery
- 발행사항
- [Sl] : The University of Wisconsin - Madison, 2025
- 발행사항
- Ann Arbor : ProQuest Dissertations & Theses, 2025
- 형태사항
- 166 p
- 주기사항
- Source: Dissertations Abstracts International, Volume: 86-12, Section: B.
- 주기사항
- Advisor: Huang, Xuhui.
- 학위논문주기
- Thesis (Ph.D.)--The University of Wisconsin - Madison, 2025.
- 초록/해제
- 요약Proteins, owing to their structural complexity and functional diversity, play a pivotal role in drug development. Recent advances in deep learning have considerably expanded the capacity to analyze protein structures, infer physicochemical properties, and model intermolecular interactions. Nevertheless, the vastness of chemical space and the intricate nature of protein-ligand interactions pose significant challenges in the early stages of drug discovery.This dissertation first addresses these challenges through the development and application of a comprehensive pipeline for fragment-based drug discovery (FBDD). A key contribution is the implementation of an automated data curation pipeline, including fetching, cleaning, and ligand decomposition tailored for ChemPLAN-Net, a deep learning framework designed to predict protein-fragment interactions using physicochemical features. This pipeline enables efficient training dataset generation and supports the systematic training ChemPLAN-Net across multiple protein families. Based on fragment predictions, virtual synthesis, screening and fragment linking are explored to develop fragment hits into candidate full ligands.One critical challenge identified during this process is the accurate prediction of fragment poses and orientations within binding pockets, which significantly impacts down stream fragment elaboration. ChemPLAN-Net, despite its efficacy in fragment identification, lacks the ability to resolve fragment posing and fragment optimization (growing, linking, and merging). To address this limitation, we introduce FeatureDock, a Transformer based deep learning model that reframes the docking problem as a spatial probability density prediction task within a more coarsely grained space, by discretizing the protein surface and learning from physicochemical features of grid points. FeatureDock introduces an interpretable probability envelope over potential binding regions, based on which compound scoring and posing estimation is possible.By integrating deep learning methodologies with protein structures and physicochemical features, this work contributes to the advancement of data-driven, interpretable, and cost-efficient drug discovery frameworks. The approaches developed herein aim to improve hit identification, reduce design-cycle latency, and ultimately increase the success rate of early-stage drug discovery pipelines.
- 일반주제명
- Chemistry
- 일반주제명
- Pharmaceutical sciences
- 일반주제명
- Biochemistry
- 키워드
- Deep learning
- 키워드
- Drug discovery
- 기타저자
- The University of Wisconsin - Madison Chemistry
- 기본자료저록
- Dissertations Abstracts International. 86-12B.
- 전자적 위치 및 접속
- 로그인 후 원문을 볼 수 있습니다.
MARC
008260126s2025 us c eng d■001000017358430
■00520260202104701
■006m o d
■007cr#unu||||||||
■020 ▼a9798280771284
■035 ▼a(MiAaPQ)AAI32116218
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a540
■1001 ▼aXue, Mingyi.
■24510▼aFeature-Based Deep Learning Approaches to Facilitate Drug Discovery
■260 ▼a[Sl]▼bThe University of Wisconsin - Madison▼c2025
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2025
■300 ▼a166 p
■500 ▼aSource: Dissertations Abstracts International, Volume: 86-12, Section: B.
■500 ▼aAdvisor: Huang, Xuhui.
■5021 ▼aThesis (Ph.D.)--The University of Wisconsin - Madison, 2025.
■520 ▼aProteins, owing to their structural complexity and functional diversity, play a pivotal role in drug development. Recent advances in deep learning have considerably expanded the capacity to analyze protein structures, infer physicochemical properties, and model intermolecular interactions. Nevertheless, the vastness of chemical space and the intricate nature of protein-ligand interactions pose significant challenges in the early stages of drug discovery.This dissertation first addresses these challenges through the development and application of a comprehensive pipeline for fragment-based drug discovery (FBDD). A key contribution is the implementation of an automated data curation pipeline, including fetching, cleaning, and ligand decomposition tailored for ChemPLAN-Net, a deep learning framework designed to predict protein-fragment interactions using physicochemical features. This pipeline enables efficient training dataset generation and supports the systematic training ChemPLAN-Net across multiple protein families. Based on fragment predictions, virtual synthesis, screening and fragment linking are explored to develop fragment hits into candidate full ligands.One critical challenge identified during this process is the accurate prediction of fragment poses and orientations within binding pockets, which significantly impacts down stream fragment elaboration. ChemPLAN-Net, despite its efficacy in fragment identification, lacks the ability to resolve fragment posing and fragment optimization (growing, linking, and merging). To address this limitation, we introduce FeatureDock, a Transformer based deep learning model that reframes the docking problem as a spatial probability density prediction task within a more coarsely grained space, by discretizing the protein surface and learning from physicochemical features of grid points. FeatureDock introduces an interpretable probability envelope over potential binding regions, based on which compound scoring and posing estimation is possible.By integrating deep learning methodologies with protein structures and physicochemical features, this work contributes to the advancement of data-driven, interpretable, and cost-efficient drug discovery frameworks. The approaches developed herein aim to improve hit identification, reduce design-cycle latency, and ultimately increase the success rate of early-stage drug discovery pipelines.
■590 ▼aSchool code: 0262.
■650 4▼aChemistry
■650 4▼aPharmaceutical sciences
■650 4▼aBiochemistry
■653 ▼aDeep learning
■653 ▼aDrug discovery
■653 ▼aFragment-based drug discovery
■653 ▼aProtein-ligand interactions
■653 ▼aVirtual screening
■690 ▼a0485
■690 ▼a0487
■690 ▼a0800
■690 ▼a0572
■71020▼aThe University of Wisconsin - Madison▼bChemistry.
■7730 ▼tDissertations Abstracts International▼g86-12B.
■790 ▼a0262
■791 ▼aPh.D.
■792 ▼a2025
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17358430▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.


