Abstract:Computing accurate reaction rates is a central challenge in computational chemistry and biology because of the high cost of free energy estimation with unbiased molecular dynamics. In this work, a data-driven machine learning algorithm is devised to learn collective variables with a multitask neural network, where a common upstream part reduces the high dimensionality of atomic configurations to a low dimensional latent space and separate downstream parts map the latent space to predictions of basin class labels and potential energies. The resulting latent space is shown to be an effective low-dimensional representation, capturing the reaction progress and guiding effective umbrella sampling to obtain accurate free energy landscapes. This approach is successfully applied to model systems including a 5D Müller Brown model, a 5D three-well model, the alanine dipeptide in vacuum, and an Au(110) surface reconstruction unit reaction. It enables automated dimensionality reduction for energy controlled reactions in complex systems, offers a unified and data-efficient framework that can be trained with limited data, and outperforms single-task learning approaches, including autoencoders.The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.jctc.1c00143.Proof showing connection between maximum likelihood methods and cross-entropy classification loss, implementation of neural networks, transition path sampling algorithm, and details of three model systems (5D Müller-Brown, 5D three-well model, alanine dipeptide, and reconstruction on Au(110) surface) (PDF)This article has not yet been cited by other publications.

Constructing Collective Variables Using Invariant Learned Representations

Understanding chemical reactions via variational autoencoder and atomic representations

Learning Effective Molecular Models from Experimental Observables.

Data-driven path collective variables

Generative deep-learning reveals collective variables of Fermionic systems

On finding optimal collective variables for complex systems by minimizing the deviation between effective and full dynamics

Multitask Machine Learning of Collective Variables for Enhanced Sampling of Rare Events

Nonlinear Intrinsic Variables and State Reconstruction in Multiscale Simulations

Descriptors-free Collective Variables From Geometric Graph Neural Networks

A unified framework for machine learning collective variables for enhanced sampling simulations: $\texttt{mlcolvar}$

Learning Collective Variables with Synthetic Data Augmentation through Physics-Inspired Geodesic Interpolation

Understanding recent deep-learning techniques for identifying collective variables of molecular dynamics

A unified framework for machine learning collective variables for enhanced sampling simulations: mlcolvar

A variational conformational dynamics approach to the selection of collective variables in metadynamics

Data-Driven Collective Variables for Enhanced Sampling

Deep Learning Collective Variables from Transition Path Ensemble

Inherent structural descriptors via machine learning

Collective Variables for Conformational Polymorphism in Molecular Crystals

Spectral Map: Embedding Slow Kinetics in Collective Variables