Litcius/Paper detail

Generative molecular dynamics

Simon Olsson

2026Current Opinion in Structural Biology6 citationsDOIOpen Access PDF

Abstract

Understanding biomolecular function depends on bridging experimental observables with models that capture structural, stationary, and dynamical properties. Molecular dynamics (MD) simulations, in principle provide a bridge, but the sampling problem remains a fundamental roadblock toward this goal. In this mini-review, I outline recent progress in the area of Generative MD (GenMD)—an approach where generative AI (GenAI) is used to mimic the statistical distributions resulting from MD simulations, which are inaccessible using current numerical algorithms. Here, I highlight a few exemplars of GenMD and then outline open problems and current limitations. • A concise and frank exposé of the state of generative artificial intelligence (AI) in molecular dynamics (MD). • Discussion of philosophical and practical contrasts between MD simulations and generative MD. • Outline of current practical limitations and roadblocks currently limiting the practical use of GenMD.

Topics & Concepts

Generative grammarBridging (networking)ObservableComputer scienceCurrent (fluid)Molecular dynamicsStatistical physicsFunction (biology)Generative modelSampling (signal processing)Artificial intelligenceDynamics (music)Uncertainty quantificationDynamical systems theoryMachine learningTheoretical computer scienceComputational modelStatistical mechanicsAlgorithmStatistical modelProtein Structure and DynamicsMachine Learning in Materials ScienceGaussian Processes and Bayesian Inference