Method write up:

For these entries a sabdab camelid nanobody was chosen essentially at random. I tried to find one that simply looked normal enough. Select amino acids on RBX1 were chosen. I chose to target either the larger alpha helix in the core, random AA’s in the undefined/unstructured tail region or, for fun, a few of the conserved cysteines in the core. Using the Old but Gold Rfantibody, rough binder backbones were created, refined with PPIflow, sequence called, refined again, and structure validated using a combo of either protenix or the venerable Boltz-2. Sorting during initial backbone generation can be summarized as a mix of target orientation screening, target contact count, target/epitope mean distance of CDR loop, and general vibes. Goofy CDR sequences got screened out manually as best able during sequence design. During refinement I implemented the recently released PPIflow, hoping to see the benefit of the maturation and refinement mechanisms but this is an ongoing work. For sorting PPIflow outputs, I mostly relied on PPIflows internal metrics, and vibes. Implausible backbones were further removed here, but I ran out of time to meaningfully screen beyond this.