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

Chamaileon is a new protein binder design framework that reframes the task as modeling a cross-context binding landscape.

It combines In-Context Complex Co-Design for training with Mixture-of-Paths Sampling at inference to generate sequences that adapt to multiple targets and states.

On the new CROSS benchmark, it improves performance across diverse conformational and multi-target settings.

The key value is enabling one sequence to work across different targets or conformations, addressing a major limitation in binder design.