I used ProtRL with ephrin B2, ephrin B3, and n425 as fixed scaffolds because their folds already resemble the general shape that interacts with the Nipah attachment protein. By keeping those structures in place, the model only explored sequence changes around the regions that face the viral protein, while the main ephrin framework stayed intact.

The model was conditioned on the Nipah attachment protein, which helped steer the generated sequences toward interfaces that looked geometrically reasonable and balanced in terms of charge and hydrophobicity. After each round of generation, I checked the predicted structures with tools that report things like pLDDT and PAE so I could see whether the fold still looked stable and whether the interface made structural sense.

I then reviewed basic biophysical signals such as how much hydrophobic surface the predictions exposed, how smoothly hydropathy shifted across the scaffold, and how the charge pattern was arranged. Designs that pushed too much hydrophobic material outward or showed charge layouts that looked unfavorable in solution tended not to pass these checks.

After going through several cycles of generating and reviewing sequences, the set gradually converged on variants that kept the scaffold’s overall structure while presenting a cleaner interface and a more balanced surface profile. The final group represents a computationally refined collection of sequences that look structurally coherent and biophysically reasonable, at least in the context of prediction.