서브메뉴
검색
Physics-Informed Deep Learning for Protein Dynamics
Physics-Informed Deep Learning for Protein Dynamics
상세정보
- 자료유형
- 학위논문 서양
- 최종처리일시
- 20260202105647
- ISBN
- 9798270233549
- DDC
- 541
- 저자명
- Liu, Bojun.
- 서명/저자
- Physics-Informed Deep Learning for Protein Dynamics
- 발행사항
- [Sl] : The University of Wisconsin - Madison, 2025
- 발행사항
- Ann Arbor : ProQuest Dissertations & Theses, 2025
- 형태사항
- 186 p
- 주기사항
- Source: Dissertations Abstracts International, Volume: 87-06, Section: B.
- 주기사항
- Advisor: Huang, Xuhui.
- 학위논문주기
- Thesis (Ph.D.)--The University of Wisconsin - Madison, 2025.
- 초록/해제
- 요약While AlphaFold2 has revolutionized protein structure prediction, it does not fully address the protein folding problem. Proteins are inherently dynamic, and uncovering the mechanisms by which they switch among distinct shapes, known as conformational changes, is crucial for understanding many biological functions. Molecular dynamics (MD) simulations provide a robust tool to study protein dynamics at atomistic details; nevertheless, extracting mechanistic insight from high-dimensional trajectories and bridging the timescale gap between femtosecond integration steps and millisecond (or longer) biological processes remain major challenges.To address this challenge, this thesis develops physics-informed deep learning frameworks that leverage principles from statistical mechanics together with modern deep learning to investigate protein dynamics. Specifically, we present GraphVAMP-nets, a geometric deep learning approach for kinetic modeling of multi-body systems, TS-DAR, an OOD-detection-inspired framework that learns meaningful hyperspherical latent representations and enables simultaneous identification of transition states across multiple free-energy barriers, and MEMnets, a deep learning method which extends Markovian dynamic models to the non-Markovian regime by explicitly incorporating memory effects for representation learning of kinetic data. Overall, this thesis opens new directions for understanding biomolecular mechanism and provides a pathway toward more effective biological interventions, including drug discovery. By integrating deep learning with physical principles, it demonstrates how elegant network design can extend applicability to complex problems in chemistry, while new theoretical insights from statistical mechanics can, in turn, guide the development of machine learning. Taken together, these developments bridge statistical mechanics and modern deep learning, establishing a unified perspective that is essential for uncovering the microscopic principles governing biomolecular systems.
- 일반주제명
- Physical chemistry
- 일반주제명
- Chemistry
- 일반주제명
- Computational chemistry
- 키워드
- Machine learning
- 기타저자
- The University of Wisconsin - Madison Chemistry
- 기본자료저록
- Dissertations Abstracts International. 87-06B.
- 전자적 위치 및 접속
- 로그인 후 원문을 볼 수 있습니다.
MARC
008260126s2025 us c eng d■001000017360980
■00520260202105647
■006m o d
■007cr#unu||||||||
■020 ▼a9798270233549
■035 ▼a(MiAaPQ)AAI32401457
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a541
■1001 ▼aLiu, Bojun.
■24510▼aPhysics-Informed Deep Learning for Protein Dynamics
■260 ▼a[Sl]▼bThe University of Wisconsin - Madison▼c2025
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2025
■300 ▼a186 p
■500 ▼aSource: Dissertations Abstracts International, Volume: 87-06, Section: B.
■500 ▼aAdvisor: Huang, Xuhui.
■5021 ▼aThesis (Ph.D.)--The University of Wisconsin - Madison, 2025.
■520 ▼aWhile AlphaFold2 has revolutionized protein structure prediction, it does not fully address the protein folding problem. Proteins are inherently dynamic, and uncovering the mechanisms by which they switch among distinct shapes, known as conformational changes, is crucial for understanding many biological functions. Molecular dynamics (MD) simulations provide a robust tool to study protein dynamics at atomistic details; nevertheless, extracting mechanistic insight from high-dimensional trajectories and bridging the timescale gap between femtosecond integration steps and millisecond (or longer) biological processes remain major challenges.To address this challenge, this thesis develops physics-informed deep learning frameworks that leverage principles from statistical mechanics together with modern deep learning to investigate protein dynamics. Specifically, we present GraphVAMP-nets, a geometric deep learning approach for kinetic modeling of multi-body systems, TS-DAR, an OOD-detection-inspired framework that learns meaningful hyperspherical latent representations and enables simultaneous identification of transition states across multiple free-energy barriers, and MEMnets, a deep learning method which extends Markovian dynamic models to the non-Markovian regime by explicitly incorporating memory effects for representation learning of kinetic data. Overall, this thesis opens new directions for understanding biomolecular mechanism and provides a pathway toward more effective biological interventions, including drug discovery. By integrating deep learning with physical principles, it demonstrates how elegant network design can extend applicability to complex problems in chemistry, while new theoretical insights from statistical mechanics can, in turn, guide the development of machine learning. Taken together, these developments bridge statistical mechanics and modern deep learning, establishing a unified perspective that is essential for uncovering the microscopic principles governing biomolecular systems.
■590 ▼aSchool code: 0262.
■650 4▼aPhysical chemistry
■650 4▼aChemistry
■650 4▼aComputational chemistry
■653 ▼aMachine learning
■653 ▼aStatistical mechanics
■653 ▼aTheoretical chemistry
■653 ▼aMolecular dynamics
■690 ▼a0494
■690 ▼a0219
■690 ▼a0485
■71020▼aThe University of Wisconsin - Madison▼bChemistry.
■7730 ▼tDissertations Abstracts International▼g87-06B.
■790 ▼a0262
■791 ▼aPh.D.
■792 ▼a2025
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17360980▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.
