Chamaileon: Cross-Context Binder Design with Contextualized Modeling and Mixed Sampling

Chamaileon presents a unified framework for designing protein binders that must work across multiple targets or conformational states. It formulates the task as cross-context binding landscape modeling, trains with In-Context Complex Co-Design (I3CD) for context-aware sequence-structure co-modeling, and uses Mixture-of-Paths Sampling (MoPS) during inference to optimize one sequence across several contexts. The authors introduce the CROSS benchmark and release code on GitHub. The abstract claims improved adaptability to diverse conformational landscapes and multi-target requirements, but detailed metrics and experimental validation are not available in the supplied material.