The ability to rapidly learn from high-dimensional data to make reliable predictions about the future of a given system is crucial in many contexts. This could be a fly avoiding predators, or the retina processing terabytes of data almost instantaneously to guide complex human actions. In this work we draw parallels between such tasks, and the efficient sampling of complex biomolecules with hundreds of thousands of atoms. For this we use the Predictive Information Bottleneck (PIB) framework developed and used for the first two classes of problems, and re-formulate it for the sampling of biomolecular structure and dynamics, especially when plagued with rare events. Our method considers a given biomolecular trajectory expressed in terms of order parameters or basis functions, and uses a deep neural network to learn the minimally complex yet most predictive aspects of this trajectory, viz the PIB. This information is used to perform iterative rounds of biased simulations that enhance the sampling along the PIB to gradually improve its accuracy, directly obtaining associated thermodynamic and kinetic information. We demonstrate the method on two test-pieces, including benzene dissociation from the protein lysozyme, where we calculate the dissociation pathway and timescales slower than milliseconds.
Copyright. All Rights Reserved.