CondenSimAdapter: A Versatile Builder for Multiscale Simulations of Protein Condensates with Broad Force-Field Compatibility and Robust Dense-Phase Relaxation

Multiscale molecular dynamics simulations that sequentially couple coarse-grained (CG) sampling with all-atom (AA) simulation are widely used to study biomolecular condensates, yet building such multiscale systems remains a practical challenge. Dense CG condensate configurations must be backmapped and converted into stable, explicitly solvated AA systems─a step where severe steric clashes often prevent production simulation, creating a “relaxation bottleneck”. Here, we introduce CondenSimAdapter, a Python package that bridges this resolution gap by integrating SE(3)-transformer–based cg2all backmapping with a physics-inspired optimization protocol (using Gaussian repulsion and soft-core potentials), which succeeds where standard energy minimization fails. CondenSimAdapter unifies four CG and nine AA force fields under a single interface. We validated the workflow by (1) demonstrating the robust elimination of major structure conflicts across diverse CG–AA combinations, (2) verifying its functional versatility in preserving the structural integrity of multidomain proteins, and (3) confirming ensemble fidelity via a 2 μs atomistic simulation of a FUS LC condensate that accurately reproduced established macroscopic and microscopic properties. By resolving the dense-phase relaxation bottleneck and providing a highly accessible, streamlined workflow, CondenSimAdapter lowers the technical barrier to multiscale condensate simulations and enables systematic, high-throughput studies of protein phase separation. CondenSimAdapter is freely available at https://github.com/hanlab-computChem/CondenSimAdapter.