Abstract
Protein complexes perform biological functions through specific structural assemblies. Although experimental techniques and AI-driven methods have resolved structures of numerous protein complexes, their assembly orders of subunits remain largely unknown. Understanding the assembly order of a protein complex is crucial for in vitro drug synthesis, subcomplex engineering, and elucidating biological mechanism. Studying the disassembly order of a protein complex may provide valuable information on its assembly order. Although experimental techniques exist to determine assembly or disassembly orders of protein complexes, they are often cost prohibitive. In this work, we address a key limitation of neglecting residue specificity of native contacts in an existing coarse-grained (CG) model. Then, a computational approach is proposed, which employs cyclic multitemperature CG simulations with reweighted strengths of native contacts, to predict disassembly orders of protein complexes. This method utilizes all-atom molecular dynamics simulations starting from the initial structure of a protein complex. By analyzing the evolution of native contacts within the molecular dynamics simulations, their residue-specific relative strengths are determined. Such information is used to re-parametrize the CG model, enabling identification of the disassembly order of the complex in CG simulations. The method has been validated using 21 protein complexes, which can predict correct disassembly orders of 14 complexes and partially correct disassembly orders of other six complexes.