Chamaileon: Cross-Context Binder Design with Contextualized Modeling and Mixed Sampling
Chamaileon is built to design one binder sequence that can work across multiple targets or protein states.
The paper frames that as “cross-context binding landscape modeling,” instead of the usual single-target, single-state setup. Its training method, In-Context Complex Co-Design, models sequence and structure with context in view. At inference, Mixture-of-Paths Sampling tries to optimize one sequence across those contexts while easing the shortage of paired multi-conformation data. The authors report gains on their CROSS benchmark for multi-state and multi-target binder requirements. HF Daily Papers' note
The paper frames that as “cross-context binding landscape modeling,” instead of the usual single-target, single-state setup. Its training method, In-Context Complex Co-Design, models sequence and structure with context in view. At inference, Mixture-of-Paths Sampling tries to optimize one sequence across those contexts while easing the shortage of paired multi-conformation data. The authors report gains on their CROSS benchmark for multi-state and multi-target binder requirements. HF Daily Papers' note
score 5