Dataset Information

MLIMC: Machine Learning-Based Implicit-Solvent Monte Carlo.

ABSTRACT: Monte Carlo (MC) methods are important computational tools for molecular structure optimizations and predictions. When solvent effects are explicitly considered, MC methods become very expensive due to the large degree of freedom associated with the water molecules and mobile ions. Alternatively implicit-solvent MC can largely reduce the computational cost by applying a mean field approximation to solvent effects and meanwhile maintains the atomic detail of the target molecule. The two most popular implicit-solvent models are the Poisson-Boltzmann (PB) model and the Generalized Born (GB) model in a way such that the GB model is an approximation to the PB model but is much faster in simulation time. In this work, we develop a machine learning-based implicit-solvent Monte Carlo (MLIMC) method by combining the advantages of both implicit solvent models in accuracy and efficiency. Specifically, the MLIMC method uses a fast and accurate PB-based machine learning (PBML) scheme to compute the electrostatic solvation free energy at each step. We validate our MLIMC method by using a benzene-water system and a protein-water system. We show that the proposed MLIMC method has great advantages in speed and accuracy for molecular structure optimization and prediction.

SUBMITTER: Chen J

PROVIDER: S-EPMC8752096 | biostudies-literature |

REPOSITORIES: biostudies-literature

ACCESS DATA

Similar Datasets

Project description:BackgroundAny Monte Carlo simulation of dose delivery using medical accelerator-generated megavolt photon beams begins by simulating electrons of the primary electron beam interacting with a target. Because the electron beam characteristics of any single accelerator are unique and generally unknown, an appropriate model of an electron beam must be assumed before MC simulations can be run. The purpose of the present study is to develop a flexible framework with suitable regression models for estimating parameters of the model of primary electron beam in simulators of medical linear accelerators using real reference dose profiles measured in a water phantom.MethodsAll simulations were run using PRIMO MC simulator. Two regression models for estimating the parameters of the simulated primary electron beam, both based on machine learning, were developed. The first model applies Principal Component Analysis to measured dose profiles in order to extract principal features of the shapes of the these profiles. The PCA-obtained features are then used by Support Vector Regressors to estimate the parameters of the model of the electron beam. The second model, based on deep learning, consists of a set of encoders processing measured dose profiles, followed by a sequence of fully connected layers acting together, which solve the regression problem of estimating values of the electron beam parameters directly from the measured dose profiles. Results of the regression are then used to reconstruct the dose profiles based on the PCA model. Agreement between the measured and reconstructed profiles can be further improved by an optimization procedure resulting in the final estimates of the parameters of the model of the primary electron beam. These final estimates are then used to determine dose profiles in MC simulations.ResultsAnalysed were a set of actually measured (real) dose profiles of 6 MV beams from a real Varian 2300 C/D accelerator, a set of simulated training profiles, and a separate set of simulated testing profiles, both generated for a range of parameters of the primary electron beam of the Varian 2300 C/D PRIMO simulator. Application of the two-stage procedure based on regression followed by reconstruction-based minimization of the difference between measured (real) and reconstructed profiles resulted in achieving consistent estimates of electron beam parameters and in a very good agreement between the measured and simulated photon beam profiles.ConclusionsThe proposed framework is a readily applicable and customizable tool which may be applied in tuning virtual primary electron beams of Monte Carlo simulators of linear accelerators. The codes, training and test data, together with readout procedures, are freely available at the site: https://github.com/taborzbislaw/DeepBeam .

Dataset Information

MLIMC: Machine Learning-Based Implicit-Solvent Monte Carlo.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets