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

Computing accurate reaction rates is a centralchallenge in computational chemistry and biology because of thehigh cost of free energy estimation with unbiased molecular dynamics.In this work, a data-driven machine learning algorithm is devised tolearn collective variables with a multitask neural network, where acommon upstream part reduces the high dimensionality of atomicconfigurations to a low dimensional latent space and separatedownstream parts map the latent space to predictions of basin classlabels and potential energies. The resulting latent space is shown to bean effective low-dimensional representation, capturing the reactionprogress and guiding effective umbrella sampling to obtain accuratefree energy landscapes. This approach is successfully applied to modelsystems including a 5D Mu''ller Brown model, a 5D three-well model, the alanine dipeptide in vacuum, and an Au(110) surfacereconstruction 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

Automated summary

This study presents a data-driven machine learning algorithm that accurately computes reaction rates. By learning collective variables and a low-dimensional latent space, it effectively captures the reaction progress and obtains accurate free energy landscapes.