Quantifying (de)Mixing of Disordered Proteins in Molecular Dynamics Simulations

Abstract

Biomolecular condensates control the spatial and temporal organization of cellular biochemistry. Their architectures often arise from complex, multicomponent mixtures governed by weak, multivalent interactions between intrinsically disordered regions (IDRs) of proteins. However, current approaches lack generalizable metrics to determine whether IDRs will mix or segregate within condensates. Here, we present a domain decomposition method (DDM) that accurately determines phase concentrations without identifying condensate interfaces. When tested against 399 IDR simulations, the DDM recovers dense- and dilute-phase concentrations within 1% agreement of established surface-based methods. By calculating spatial variances to map three-dimensional organization, this approach successfully distinguishes homogeneous mixing from surface enrichment, core--shell architectures, and complete segregation. Applying this continuous metric to 2,068 binary mixtures of 32 distinct IDRs, we demonstrate that physicochemical similarity strongly promotes mixing. Hydrophobic IDRs mixed promiscuously across all partner chemistries, whereas charged sequences mixed in a way highly sensitive to charge complementarity, forming the most stable condensates as net charge approached zero. Compared with experimental networks, our IDR-only simulations predicted the partitioning preferences for 10 of 13 proteins, with discrepancies accurately isolating cases where colocalization strictly requires RNA or folded domains. Quantifying mixing in this way provides a robust and reproducable method for predicting IDR interactomes and understanding the composition of complex cellular condensates.