Description
This submission uses a de novo protein binder design pipeline targeting the RBX1 RING domain, with emphasis on the E2-interaction surface rather than the zinc-coordinating face. We generated compact, mostly helical binder backbones with RFdiffusion, designed sequences with zinc-aware LigandMPNN using amino acid biases, filtered candidates for sequence quality and monomer foldability, and then ranked and selected candidates using AlphaFold 3 interface confidence, with ipSAE as the primary selection metric.
My central hypothesis is that compact binders engaging the E2-facing surface of RBX1 while avoiding zinc-face artifacts are more likely to produce stable and experimentally tractable binders. Final sequences were selected using a gate-based process that prioritized predicted interface quality, low disorder and clash risk, and diversity across distinct backbones, rather than over-optimizing a single score. This submission therefore reflects a balance of interface confidence, foldability, and structural diversity.
