Protein binder design is a crucial task in structural biology, but existing methods typically assume a single target in a single structural state. This limits their applicability to real-world scenarios where a binder must function across multiple conformations or targets. To address this, researchers have developed Chamaileon, a unified framework for cross-context binder design.
Chamaileon is built on two key innovations: In-Context Complex Co-Design (I3CD) for context-aware sequence-structure modeling, and Mixture-of-Paths Sampling (MoPS), which iteratively optimizes a shared sequence across multiple structural contexts during inference. This allows the generation of a single binder sequence that can adapt its structure to satisfy distinct binding contexts.
The team also introduced CROSS, a benchmark covering both multi-state and multi-target binder design. Extensive evaluation on CROSS demonstrates that Chamaileon effectively generates sequences adaptable to diverse conformational landscapes and multi-target requirements, opening a path toward programmable multi-specific binders and conformational modulators.
The code is available on GitHub at https://github.com/caohengyuan/Chamaileon.