RADiAnce is a novel framework for designing protein complexes targeting specific sites. This model leverages existing interfaces to guide the design of novel complexes, integrating search and generation within a common contrast latent space. This allows for efficient identification of relevant interfaces and seamless integration via a conditional latent diffusion generator, enabling cross-domain interface transfer. RADiAnce outperforms existing models across various metrics and demonstrates cross-domain generalization, demonstrating that searching for interfaces across multiple domains can improve the performance of complex generation across other domains.
