HEroBM: a deep equivariant graph neural network for high-fidelity backmapping from coarse-grained to all-atom structures

Molecular simulations play a pivotal role in chemistry, biology, and material sciences, enabling thestudy of complex dynamic properties within systems. Coarse-grained (CG) techniques have emergedas indispensable tools in this domain, facilitating the sampling of large-scale systems and extendingsimulation timescales by simplifying system representation. However, CG approaches involve a trade-off: they sacrifice atomistic details that may be crucial for understanding the underlying processes.To address this challenge, a recommended strategy is to identify key CG conformations and employbackmapping methods to retrieve atomistic coordinates. Currently, rule-based methods often yieldsuboptimal geometries and rely on energy relaxation, resulting in less-than-optimal outcomes. Incontrast, machine learning techniques offer higher accuracy but may lack transferability betweensystems or be tied to specific CG mappings. In this study, we present HEroBM, a dynamic and scalablemethod that utilizes deep equivariant graph neural networks and a hierarchical approach to achievehigh-resolution backmapping. HEroBM is capable of handling any type of CG mapping, providing aversatile and efficient protocol for reconstructing atomistic structures with high accuracy. Groundedin local principles, HEroBM spans the entire chemical space and can be applied across systems ofvarying composition and sizes. We demonstrate the versatility of our framework through a range ofbiological systems, including a complex real-case scenario. Here, our end-to-end backmapping approachaccurately generates atomistic coordinates for a G protein-coupled receptor bound to an organic smallmolecule within a cholesterol/phospholipid bilayer. The high-fidelity HEroBM backmapping enablesresearchers to effortlessly transition between CG and all-atom simulations, opening unprecedentedavenues for molecular investigations.