Molecular dynamics (MD) simulations are a cornerstone of computational biophysics, providing atomistic insight into the conformational dynamics of complex biomolecules. Despite their predictive power, the reliability of MD simulations is hindered by two key limitations: limited sampling of longtimescale events and inaccuracies in the underlying force fields. To ensure meaningful predictions, simulations must therefore be rigorously validated and refined against experimental data. This thesis addresses these challenges by advancing the integration of experimental measurements into MD simulations within a unified, Bayesianinspired framework. Building on recent efforts to simultaneously refine structural ensembles and forward models, we extend the scope to include force-field corrections, enabling a synergistic optimization that improves both sampling consistency and force-field accuracy. To this aim, we design and implemented MDRefine, a Python package that provides a flexible platform for the refinement of ensembles, force fields, and forward models within a single workflow. The package incorporates tunable hyperparameters to balance experimental constraints and prior knowledge, and its effectiveness is demonstrated in benchmark applications to RNA oligomers. Furthermore, we revisit ensemble refinement from a fully Bayesian perspective, moving beyond maximum-a-posteriori estimates to emphasize the importance of sampling the full posterior distribution. In this context, we identify a fundamental normalization issue in existing approaches and resolve it by introducing the Jeffreys uninformative prior, which has been implemented in MDRefine. The methodological advances presented in this thesis contribute to a more robust integration of simulations and experiments, advancing the efforts toward more accurate MD simulations. Applications to nucleic acids demonstrate both the practical utility of the framework and its potential to be generalized to diverse molecular systems.
Development and application of methods to integrate molecular simulations with experimental measurements
GILARDONI, IVAN
2025
Abstract
Molecular dynamics (MD) simulations are a cornerstone of computational biophysics, providing atomistic insight into the conformational dynamics of complex biomolecules. Despite their predictive power, the reliability of MD simulations is hindered by two key limitations: limited sampling of longtimescale events and inaccuracies in the underlying force fields. To ensure meaningful predictions, simulations must therefore be rigorously validated and refined against experimental data. This thesis addresses these challenges by advancing the integration of experimental measurements into MD simulations within a unified, Bayesianinspired framework. Building on recent efforts to simultaneously refine structural ensembles and forward models, we extend the scope to include force-field corrections, enabling a synergistic optimization that improves both sampling consistency and force-field accuracy. To this aim, we design and implemented MDRefine, a Python package that provides a flexible platform for the refinement of ensembles, force fields, and forward models within a single workflow. The package incorporates tunable hyperparameters to balance experimental constraints and prior knowledge, and its effectiveness is demonstrated in benchmark applications to RNA oligomers. Furthermore, we revisit ensemble refinement from a fully Bayesian perspective, moving beyond maximum-a-posteriori estimates to emphasize the importance of sampling the full posterior distribution. In this context, we identify a fundamental normalization issue in existing approaches and resolve it by introducing the Jeffreys uninformative prior, which has been implemented in MDRefine. The methodological advances presented in this thesis contribute to a more robust integration of simulations and experiments, advancing the efforts toward more accurate MD simulations. Applications to nucleic acids demonstrate both the practical utility of the framework and its potential to be generalized to diverse molecular systems.File | Dimensione | Formato | |
---|---|---|---|
PhD_Thesis.pdf
embargo fino al 24/09/2026
Dimensione
5.4 MB
Formato
Adobe PDF
|
5.4 MB | Adobe PDF |
I documenti in UNITESI sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/20.500.14242/296416
URN:NBN:IT:SISSA-296416