ESMAdam: a plug-and-play all-purpose protein ensemble generator

Abstract

Proteins often adopt multiple ensemble conformations to perform essential functions such as catalysis, transport, and signal transduction. Traditional physics-based methods for generating these conformations, including molecular dynamics and Monte Carlo simulations, are computationally expensive and time-consuming, limiting their practicality for high-throughput applications like screening. Recent advances in machine learning, particularly deep generative models, offer a promising alternative for protein conformation ensemble generation. However, these models are often task-specific or rely on strong assumptions to generalize. Here, we introduce ESMAdam, a versatile and efficient framework for protein conformation ensemble generation. Using the ESMFold protein language model ESMFold and ADAM stochastic optimization in the continuous protein embedding space, ESMAdam addresses a wide range of ensemble generation tasks. In this work, we demonstrate several basic applications of ESMAdam, including conditional ensemble generation and CG-to-all-atom backmapping. In addition, we showcase advanced applications, such as screening alternative binding modes of protein multimers and reconstructing 3D structures from cryo-EM images. Compared to traditional physics-based methods, ESMAdam significantly reduces computational time. Unlike deep-generative-model-based approaches, it requires no retraining and easily adapts to diverse ensemble restraint conditions, making it exceptionally suited for various structure prediction and screening tasks. This plug-and-play framework represents a step toward efficient and flexible protein ensemble generation for applications in structural biology and drug discovery.