While structure-based protein design has accelerated the generation of novel complexes, designing complexes across large or multidomain interfaces remains challenging due to high computational costs and declining success rates with increasing target size. In this study, we hypothesize that protein folding neural networks (PFNNs) operate in a "local-first" manner, prioritizing local interactions and limiting sensitivity to overall folding potential. Based on this hypothesis, we propose an epitope-specific strategy that retains only discrete surface residues surrounding the binding site. This approach improves in silico success rates by up to 80% and reduces the average time per successful design by up to 40x compared to full-domain workflows, enabling complex design for previously challenging targets such as ClpP and ALS3. Furthermore, we developed a customized pipeline that includes a Monte Carlo-based evolution step to overcome local minima and a site-specific biased defolding step to refine sequence patterns. These advances not only establish a generalizable framework for efficient fusion design for structurally large and inaccessible targets, but also support the "local-first" hypothesis as a guiding principle for PFNN-based designs.
