Summary

We explore repeat-protein folds as starting points for de novo binder design. The rationale is that many natural repeat proteins mediate protein–protein interactions, are modular and autonomously foldable, and therefore provide stable backbones for engineering. A canonical example is the DARPin scaffold, which is derived from ankyrin repeats. More broadly, repeat architectures such as ankyrin, HEAT-like, and leucine-rich repeat (LRR) proteins offer curved, adaptable binding surfaces that can engage a range of target geometries. In addition to these favourable biophysical properties, repeat domains are well represented in structural databases. Our submissions span several repeat-protein superfamilies and topologies: ankyrin repeats (CATH 1.25.40.20), classic LRR (CATH 3.80.10), and LRR variants (CATH 1.25.10.10).

Design methodology

The primary tools used to design our submissions are BoltzGen and ProteinMPNN. We use these methods to explore several paths through the design process, which can be summarised broadly as follows:

1. De novo backbone generation with BoltzGen. BoltzGen is conditioned on a template of RBX1 (modelled as a standalone RING domain including 3 zinc ions) together with binder scaffold geometry provided via templates of one of the three repeat scaffolds described above. These scaffolds have no natural affinity to RBX1, and as such, designs are fully de novo. Boltz-IF is called on each generated backbone to produce an initial pool of sequence candidates, and backbones are ranked by the median log-likelihood of their pool. The aim is to identify binder geometries that allow high-quality sequence exploration. A subset of top backbone candidates from each scaffold track is taken forward.
2. Sequence screen. Each backbone candidate enters a larger sequence-space screen using either ProteinMPNN (LRR variant track) or Boltz-IF (classic LRR and ankyrin repeat tracks). Top candidates are ranked by log-likelihood and taken forward.
3. Refolding and self-consistency. Top sequence–backbone candidates are refolded using Boltz2 to determine whether each design recovers its intended binding mode. We extract ML-based confidence scores from Boltz2 and select designs that maximise both binder confidence (e.g. pTM, pLDDT) and interface quality (e.g. ipTM, ipSAE), filtering out designs that fail self-consistency checks across multiple diffusion samples or dock in unintended orientations.

Sequences in this submission represent de novo designs generated using the classic LRR scaffold.