서브메뉴
검색
Modeling Dynamics of Multi-Body Systems via Machine Learning and Non-Markovian Approaches
Modeling Dynamics of Multi-Body Systems via Machine Learning and Non-Markovian Approaches
상세정보
- 자료유형
- 학위논문 서양
- 최종처리일시
- 20250211152938
- ISBN
- 9798384458180
- DDC
- 541
- 저자명
- Qiu, Yunrui.
- 서명/저자
- Modeling Dynamics of Multi-Body Systems via Machine Learning and Non-Markovian Approaches
- 발행사항
- [Sl] : The University of Wisconsin - Madison, 2024
- 발행사항
- Ann Arbor : ProQuest Dissertations & Theses, 2024
- 형태사항
- 217 p
- 주기사항
- Source: Dissertations Abstracts International, Volume: 86-04, Section: B.
- 주기사항
- Advisor: Huang, Xuhui.
- 학위논문주기
- Thesis (Ph.D.)--The University of Wisconsin - Madison, 2024.
- 초록/해제
- 요약Multi-body systems are widely prevalent in chemistry, biology, and material sciences. Their complex energy landscapes, heterogeneous dynamics across different time-scales, and numerous pathways pose significant challenges in modeling long-term dynamics and understanding the underlying molecular mechanisms with high spatial and temporal resolution using current experimental and computational techniques. In this thesis, we developed machine learning algorithms and non-Markovian dynamics modeling approaches to tackle these challenges and explore the dynamics of multi-body systems, ranging from biomolecules, such as protein-protein encounter complexes and chromatin, to materials like supercooled liquids. In particular, to bridge the time gap between simulations and the heterogeneous dynamics of interest, and to provide better interpretation for the underlying mechanisms, we developed a non-Markovian dynamic modeling approach called the Integrated Generalized Master Equation (IGME) model. Unlike conventional Markov State Models (MSMs), the IGME model encodes non-Markovian dynamics into time-integration of memory kernel functions and offers more accurate predictions for long-time dynamics based on shorter simulations. Additionally, to categorize diverse pathways with comparable fluxes, we designed the Latent-space Path Clustering (LPC) algorithm, which applies variational autoencoder network to effectively classify multiple pathways into a small set of metastable path channels according to their kinetic similarities and path typologies. Moreover, we have developed an information bottleneck approach for MSM constructions, providing an end-to-end pipeline that achieves state-of-the-art performance. With these effective machine learning and dynamic modeling tools, we studied a protein-protein encounter complex system, where our IGME model successfully predicted multiple non-canonical metastable protein-protein interfaces, supporting the rational design of PROTACs, a promising next-generation cancer treatment drug. Meanwhile, using our LPC algorithm and IGME model, we also explored chromatin folding dynamics and mechanisms, examining the effects of phase separation of nucleosome condensation and DNA linker length, providing insights into the discrepancies between in vivo and in vitro studies. In addition, we developed an unsupervised time-lagged approach to efficiently uncover the structural origins of dynamical heterogeneities in multi-body supercooled liquids, addressing a key open question in the field in a much more data-efficient manner.
- 일반주제명
- Physical chemistry
- 일반주제명
- Computational chemistry
- 일반주제명
- Biophysics
- 일반주제명
- Systematic biology
- 일반주제명
- Genetics
- 키워드
- Drug discovery
- 키워드
- Machine learning
- 기타저자
- The University of Wisconsin - Madison Chemistry
- 기본자료저록
- Dissertations Abstracts International. 86-04B.
- 전자적 위치 및 접속
- 로그인 후 원문을 볼 수 있습니다.
MARC
008250123s2024 us c eng d■001000017164249
■00520250211152938
■006m o d
■007cr#unu||||||||
■020 ▼a9798384458180
■035 ▼a(MiAaPQ)AAI31564347
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a541
■1001 ▼aQiu, Yunrui.
■24510▼aModeling Dynamics of Multi-Body Systems via Machine Learning and Non-Markovian Approaches
■260 ▼a[Sl]▼bThe University of Wisconsin - Madison▼c2024
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2024
■300 ▼a217 p
■500 ▼aSource: Dissertations Abstracts International, Volume: 86-04, Section: B.
■500 ▼aAdvisor: Huang, Xuhui.
■5021 ▼aThesis (Ph.D.)--The University of Wisconsin - Madison, 2024.
■520 ▼aMulti-body systems are widely prevalent in chemistry, biology, and material sciences. Their complex energy landscapes, heterogeneous dynamics across different time-scales, and numerous pathways pose significant challenges in modeling long-term dynamics and understanding the underlying molecular mechanisms with high spatial and temporal resolution using current experimental and computational techniques. In this thesis, we developed machine learning algorithms and non-Markovian dynamics modeling approaches to tackle these challenges and explore the dynamics of multi-body systems, ranging from biomolecules, such as protein-protein encounter complexes and chromatin, to materials like supercooled liquids. In particular, to bridge the time gap between simulations and the heterogeneous dynamics of interest, and to provide better interpretation for the underlying mechanisms, we developed a non-Markovian dynamic modeling approach called the Integrated Generalized Master Equation (IGME) model. Unlike conventional Markov State Models (MSMs), the IGME model encodes non-Markovian dynamics into time-integration of memory kernel functions and offers more accurate predictions for long-time dynamics based on shorter simulations. Additionally, to categorize diverse pathways with comparable fluxes, we designed the Latent-space Path Clustering (LPC) algorithm, which applies variational autoencoder network to effectively classify multiple pathways into a small set of metastable path channels according to their kinetic similarities and path typologies. Moreover, we have developed an information bottleneck approach for MSM constructions, providing an end-to-end pipeline that achieves state-of-the-art performance. With these effective machine learning and dynamic modeling tools, we studied a protein-protein encounter complex system, where our IGME model successfully predicted multiple non-canonical metastable protein-protein interfaces, supporting the rational design of PROTACs, a promising next-generation cancer treatment drug. Meanwhile, using our LPC algorithm and IGME model, we also explored chromatin folding dynamics and mechanisms, examining the effects of phase separation of nucleosome condensation and DNA linker length, providing insights into the discrepancies between in vivo and in vitro studies. In addition, we developed an unsupervised time-lagged approach to efficiently uncover the structural origins of dynamical heterogeneities in multi-body supercooled liquids, addressing a key open question in the field in a much more data-efficient manner.
■590 ▼aSchool code: 0262.
■650 4▼aPhysical chemistry
■650 4▼aComputational chemistry
■650 4▼aBiophysics
■650 4▼aSystematic biology
■650 4▼aGenetics
■653 ▼aChromatin nucleosomes
■653 ▼aDrug discovery
■653 ▼aMachine learning
■653 ▼aMulti-body dynamical systems
■653 ▼aNon-Markovian dynamics model
■653 ▼aSupercooled liquids
■690 ▼a0494
■690 ▼a0219
■690 ▼a0786
■690 ▼a0423
■690 ▼a0800
■690 ▼a0369
■71020▼aThe University of Wisconsin - Madison▼bChemistry.
■7730 ▼tDissertations Abstracts International▼g86-04B.
■790 ▼a0262
■791 ▼aPh.D.
■792 ▼a2024
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17164249▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.
