A language model-informed minimalistic design of functional coacervate vesicles

Hierarchical biopolymer interactions in nature yield intricate architectures, while synthetic biomimetics adopt a reductionist approach to create scalable and tunable systems. Inspired by intracellular liquid-liquid phase separation into membraneless, vesicle-like condensates, synthetic coacervate vesicles (CVs) have emerged for drug delivery, protocell design, and catalysis. However, a general reductionist framework for rationally constructing CVs remains challenging due to insufficient understanding of the minimal motifs required to drive coacervates into stable vesicular topologies. Here, we present a protein language model-informed design of such minimal systems, where arginine-rich oligopeptides and competing anionic oligomers self-assemble into oligo-coacervate vesicles (OCVs) without any chemical modifications or external fuels. Combining experiments and molecular dynamics simulations, we demonstrate an assembly mechanism governed by an electrostatic-hydrophobic balance, which enables vesiculation across a broad range of oligomer pairs. This principle further permits de novo functionalization, exemplified by OCVs engineered to selectively recognize and encapsulate DNA G-quadruplex structures. Together, these findings establish sequence-encoded design rules for the predictable construction of vesicle-like compartments. The study presents a protein language model guided design of minimal coacervate vesicles that are composed entirely of oligomeric biomolecules, with functions programmable through oligopeptide sequences.