Machine learning in multiscale modeling and simulations of molecular systems

Collective variables (CVs) are low-dimensional representations of the state of a complex system, which help us rationalize molecular conformations and sample free energy landscapes with molecular dynamics simulations. However, identifying a representative set of CVs for a given system is far from ob...

Descripción completa

Detalles Bibliográficos
Autor: Hashemian, Behrooz
Tipo de recurso: tesis doctoral
Fecha de publicación:2015
País:España
Institución:Universitat Politècnica de Catalunya (UPC)
Repositorio:UPCommons. Portal del coneixement obert de la UPC
Idioma:inglés
OAI Identifier:oai:upcommons.upc.edu:2117/95711
Acceso en línea:https://hdl.handle.net/2117/95711
https://dx.doi.org/10.5821/dissertation-2117-95711
Access Level:acceso abierto
Palabra clave:Modelització en etapes múltiples
Dinàmica molecular -- Mètodes de simulació
Àrees temàtiques de la UPC::Matemàtiques i estadística
id ES_025f285af8901bb7bb41e5fed9aa2e35
oai_identifier_str oai:upcommons.upc.edu:2117/95711
network_acronym_str ES
network_name_str España
repository_id_str
dc.title.none.fl_str_mv Machine learning in multiscale modeling and simulations of molecular systems
title Machine learning in multiscale modeling and simulations of molecular systems
spellingShingle Machine learning in multiscale modeling and simulations of molecular systems
Hashemian, Behrooz
Modelització en etapes múltiples
Dinàmica molecular -- Mètodes de simulació
Àrees temàtiques de la UPC::Matemàtiques i estadística
title_short Machine learning in multiscale modeling and simulations of molecular systems
title_full Machine learning in multiscale modeling and simulations of molecular systems
title_fullStr Machine learning in multiscale modeling and simulations of molecular systems
title_full_unstemmed Machine learning in multiscale modeling and simulations of molecular systems
title_sort Machine learning in multiscale modeling and simulations of molecular systems
dc.creator.none.fl_str_mv Hashemian, Behrooz
author Hashemian, Behrooz
author_facet Hashemian, Behrooz
author_role author
dc.contributor.none.fl_str_mv Arroyo Balaguer, Marino
dc.subject.none.fl_str_mv Modelització en etapes múltiples
Dinàmica molecular -- Mètodes de simulació
Àrees temàtiques de la UPC::Matemàtiques i estadística
topic Modelització en etapes múltiples
Dinàmica molecular -- Mètodes de simulació
Àrees temàtiques de la UPC::Matemàtiques i estadística
description Collective variables (CVs) are low-dimensional representations of the state of a complex system, which help us rationalize molecular conformations and sample free energy landscapes with molecular dynamics simulations. However, identifying a representative set of CVs for a given system is far from obvious, and most often relies on physical intuition or partial knowledge about the systems. An inappropriate choice of CVs is misleading and can lead to inefficient sampling. Thus, there is a need for systematic approaches to effectively identify CVs. In recent years, machine learning techniques, especially nonlinear dimensionality reduction (NLDR), have shown their ability to automatically identify the most important collective behavior of molecular systems. These methods have been widely used to visualize molecular trajectories. However, in general they do not provide a differentiable mapping from high-dimensional configuration space to their low-dimensional representation, as required in enhanced sampling methods, and they cannot deal with systems with inherently nontrivial conformational manifolds. In the fist part of this dissertation, we introduce a methodology that, starting from an ensemble representative of molecular flexibility, builds smooth and nonlinear data-driven collective variables (SandCV) from the output of nonlinear manifold learning algorithms. We demonstrate the method with a standard benchmark molecule and show how it can be non-intrusively combined with off-the-shelf enhanced sampling methods, here the adaptive biasing force method. SandCV identifies the system's conformational manifold, handles out-of-manifold conformations by a closest point projection, and exactly computes the Jacobian of the resulting CVs. We also illustrate how enhanced sampling simulations with SandCV can explore regions that were poorly sampled in the original molecular ensemble. We then demonstrate that NLDR methods face serious obstacles when the underlying CVs present periodicities, e.g.~arising from proper dihedral angles. As a result, NLDR methods collapse very distant configurations, thus leading to misinterpretations and inefficiencies in enhanced sampling. Here, we identify this largely overlooked problem, and discuss possible approaches to overcome it. Additionally, we characterize flexibility of alanine dipeptide molecule and show that it evolves around a flat torus in four-dimensional space. In the final part of this thesis, we propose a novel method, atlas of collective variables, that systematically overcomes topological obstacles, ameliorates the geometrical distortions and thus allows NLDR techniques to perform optimally in molecular simulations. This method automatically partitions the configuration space and treats each partition separately. Then, it connects these partitions from the statistical mechanics standpoint.
publishDate 2015
dc.date.none.fl_str_mv 2015
2015-05-08
2015
2015-07-16
dc.type.none.fl_str_mv doctoral thesis
http://purl.org/coar/resource_type/c_db06
VoR
http://purl.org/coar/version/c_970fb48d4fbd8a85
dc.type.openaire.fl_str_mv info:eu-repo/semantics/doctoralThesis
format doctoralThesis
dc.identifier.none.fl_str_mv https://hdl.handle.net/2117/95711
https://dx.doi.org/10.5821/dissertation-2117-95711
url https://hdl.handle.net/2117/95711
https://dx.doi.org/10.5821/dissertation-2117-95711
dc.language.none.fl_str_mv Inglés
eng
language_invalid_str_mv Inglés
language eng
dc.rights.none.fl_str_mv open access
http://purl.org/coar/access_right/c_abf2
dc.rights.openaire.fl_str_mv info:eu-repo/semantics/openAccess
rights_invalid_str_mv open access
http://purl.org/coar/access_right/c_abf2
eu_rights_str_mv openAccess
dc.format.none.fl_str_mv application/pdf
dc.publisher.none.fl_str_mv Universitat Politècnica de Catalunya
publisher.none.fl_str_mv Universitat Politècnica de Catalunya
dc.source.none.fl_str_mv reponame:UPCommons. Portal del coneixement obert de la UPC
instname:Universitat Politècnica de Catalunya (UPC)
instname_str Universitat Politècnica de Catalunya (UPC)
reponame_str UPCommons. Portal del coneixement obert de la UPC
collection UPCommons. Portal del coneixement obert de la UPC
repository.name.fl_str_mv
repository.mail.fl_str_mv
_version_ 1869402650317946880
spelling Machine learning in multiscale modeling and simulations of molecular systemsHashemian, BehroozModelització en etapes múltiplesDinàmica molecular -- Mètodes de simulacióÀrees temàtiques de la UPC::Matemàtiques i estadísticaCollective variables (CVs) are low-dimensional representations of the state of a complex system, which help us rationalize molecular conformations and sample free energy landscapes with molecular dynamics simulations. However, identifying a representative set of CVs for a given system is far from obvious, and most often relies on physical intuition or partial knowledge about the systems. An inappropriate choice of CVs is misleading and can lead to inefficient sampling. Thus, there is a need for systematic approaches to effectively identify CVs. In recent years, machine learning techniques, especially nonlinear dimensionality reduction (NLDR), have shown their ability to automatically identify the most important collective behavior of molecular systems. These methods have been widely used to visualize molecular trajectories. However, in general they do not provide a differentiable mapping from high-dimensional configuration space to their low-dimensional representation, as required in enhanced sampling methods, and they cannot deal with systems with inherently nontrivial conformational manifolds. In the fist part of this dissertation, we introduce a methodology that, starting from an ensemble representative of molecular flexibility, builds smooth and nonlinear data-driven collective variables (SandCV) from the output of nonlinear manifold learning algorithms. We demonstrate the method with a standard benchmark molecule and show how it can be non-intrusively combined with off-the-shelf enhanced sampling methods, here the adaptive biasing force method. SandCV identifies the system's conformational manifold, handles out-of-manifold conformations by a closest point projection, and exactly computes the Jacobian of the resulting CVs. We also illustrate how enhanced sampling simulations with SandCV can explore regions that were poorly sampled in the original molecular ensemble. We then demonstrate that NLDR methods face serious obstacles when the underlying CVs present periodicities, e.g.~arising from proper dihedral angles. As a result, NLDR methods collapse very distant configurations, thus leading to misinterpretations and inefficiencies in enhanced sampling. Here, we identify this largely overlooked problem, and discuss possible approaches to overcome it. Additionally, we characterize flexibility of alanine dipeptide molecule and show that it evolves around a flat torus in four-dimensional space. In the final part of this thesis, we propose a novel method, atlas of collective variables, that systematically overcomes topological obstacles, ameliorates the geometrical distortions and thus allows NLDR techniques to perform optimally in molecular simulations. This method automatically partitions the configuration space and treats each partition separately. Then, it connects these partitions from the statistical mechanics standpoint.Las variables colectivas (CVs, acrónimo inglés de collective variables) son representaciones de baja dimensionalidad del estado de un sistema complejo, que nos ayudan a racionalizar conformaciones moleculares y muestrear paisajes de energía libre con simulaciones de dinámica molecular. Sin embargo, identificar unas CVs representativas para un sistema dado dista de ser evidente, por lo que a menudo se confía en la intuición física o en el conocimiento parcial de los sistemas bajo estudio. Una elección inadecuada de las CVs puede dar a interpretaciones engañosas y conducir a un muestreo ineficiente. Por lo tanto, hay una necesidad de desarrollar enfoques sistemáticos para identificar CVs de manera efectiva. En los últimos años, las técnicas de aprendizaje de máquina, especialmente las técnicas de reducción de dimensionalidad no lineal (NLDR, acrónimo inglés de nonlinear dimensionality reduction), han demostrado su capacidad para identificar automáticamente el comportamiento colectivo de sistemas moleculares. Estos métodos han sido ampliamente utilizados para visualizar las trayectorias moleculares. No obstante, en general las técnicas de NLDR no proporcionan una aplicación diferenciable de las configuraciones de alta dimensión a su representación de baja dimensión, condición que es requerida en los métodos mejorados de muestreo, por lo que no pueden hacer frente a sistemas con variedades conformacionales inherentemente no triviales. En la primer parte de esta tesis doctoral, introducimos una metodología que, a partir de un conjunto de conformaciones representativo de la flexibilidad del sistema molecular, construye variables colectivas suaves y no lineales basadas en datos (SandCV, acrónimo en inglés de smooth and nonlinear data-driven collective variables) obtenidos utilizando algoritmos de aprendizaje de variedades no lineales. Demostramos el método con una molécula de referencia estándar y mostramos cómo puede ser combinado de forma no intrusiva con métodos mejorados de muestreo ya existentes, aquí el método de la fuerza de sesgo adaptativa. SandCV identifica la variedad conformacional del sistema, maneja conformaciones fuera de la variedad por una proyección al punto más cercano de la variedad, y calcula exactamente el Jacobiano de las CVs resultantes. También ilustramos cómo simulaciones de muestreo mejoradas pueden, mediante SandCV, explorar regiones que fueron mal muestreadas en el conjunto molecular inicial. A continuación, demostramos que los métodos NLDR se enfrentan a serios obstáculos cuando las CVs subyacentes presentan periodicidad, por ejemplo, derivados de ángulos diedrales. Como consecuencia, los métodos NLDR colapsan configuraciones muy distantes, lo que conduce a interpretaciones erróneas y a ineficiencias en el muestreo mejorado. Aquí, identificamos este problema en gran medida pasado por alto, y discutimos los posibles enfoques para superarlo. Además, caracterizamos la flexibilidad de la molécula de dipéptido alanina y demostramos que evoluciona en torno a un toro plano en cuatro dimensiones. En la parte final de esta tesis, proponemos una metodología novedosa, atlas de variables colectivas, que supera sistemáticamente obstáculos topológicos, aminora las distorsiones geométricas y por lo tanto permite que las técnicas NLDR trabajen de manera óptima en simulaciones moleculares. Este método divide de forma automática el espacio configuracional y trata a cada partición por separado. Después, conecta estas particiones del punto de vista de mecánica estadística.Universitat Politècnica de CatalunyaArroyo Balaguer, Marino20152015-05-0820152015-07-16doctoral thesishttp://purl.org/coar/resource_type/c_db06VoRhttp://purl.org/coar/version/c_970fb48d4fbd8a85info:eu-repo/semantics/doctoralThesisapplication/pdfhttps://hdl.handle.net/2117/95711https://dx.doi.org/10.5821/dissertation-2117-95711reponame:UPCommons. Portal del coneixement obert de la UPCinstname:Universitat Politècnica de Catalunya (UPC)Inglésengopen accesshttp://purl.org/coar/access_right/c_abf2info:eu-repo/semantics/openAccessoai:upcommons.upc.edu:2117/957112026-05-27T15:37:01Z
score 15,301603

Similares en LA Referencia