Method

1. Target Region Analysis

We estimated the beta-sheet propensity of the RBX1 target using LS2P [1], identifying residues 21-35 as a high beta-sheet propensity region suitable for strand-mediated protein-protein interactions.

1. Structure Generation with RFdiffusion

We used RFdiffusion [2] to generate binder backbones that form beta-strand contacts with the target region. Key parameters:

• 100-residue binder with full RBX1 (residues 1-108) as context
• inpaint_str_strand targeting residues A21-35 to enforce beta-sheet formation at the interface
• All residues 21-35 specified as PPI hotspot residues
• 1,000 designs per scaffold seed across multiple seeds

1. Sequence Design with SolubleMPNN

Binder backbone structures were passed to SolubleMPNN [3] to generate sequences optimized for solubility and foldability while maintaining the designed interface contacts.

1. In Silico Validation with Chai-1

Each designed binder-RBX1 complex was predicted using Chai-1 [4] with 5 model seeds per design. We evaluated two criteria: a. Beta-sheet formation: Secondary structure assignment (DSSP) was used to verify that the binder forms a beta-sheet at the target interface region across all 5 models. b. Interface quality: The interface predicted TM-score (ipTM) was used as a confidence metric for binding.

1. Ranking

Binders were ranked by:

1. Requiring beta-sheet formation at the target region in all 5 Chai-1 models (52,084 / 257,409 designs passed)
2. Sorting by mean ipTM across the 5 models (top-1 mean ipTM = 0.878)

The top 100 binders were selected for submission.

---

In Silico Results

• 257,409 total binder designs were evaluated
• 52,084 (20.2%) formed beta-sheets at the target region in all 5 Chai-1 models
• Top 100 binders had mean ipTM ranging from 0.878 to 0.845