We generated a large and diverse panel of de novo nanobody binders targeting the Nipah virus using the nanobody-anything protocol implemented in BoltzGen. BoltzGen enables high-throughput sequence–structure co-design but does not rank candidates using iPTM or iPSAE like many other tools making it a unique tool. To ensure that only the most structurally plausible and target-engaged designs progressed, we implemented a two-stage refinement and filtering pipeline. From an initial pool of ~30,000 BoltzGen-generated nanobodies, we selected the top 400 candidates based on BoltzGen implemented filter. Then ,these candidates were re-folded against the Nipah target using Boltz2, employing a more stringent structural prediction regime optimized for loop-mediated interactions. Because nanobody–antigen binding is often dominated by long, flexible CDR loops—and because loop convergence improves with deeper recycling—we ran Boltz2 with 20 recycles and extracted three structural predictions per design. This higher-recycle protocol follows established recommendations for improving accuracy in loop-dominated interfaces, providing enhanced confidence in binder geometry and epitope engagement.

The final set of binders were ranked by ipSAE (which when observed for 20x recycles is higher than what is listed on here, likely due to less number of cycles). This is additionally useful to compare BoltzGen derived specific ranking for choosing top hits based on their reported quality score which considered several parameters, while here we select top 400 candidates that BoltzGen ranks and re-fold and select top-ipSAE.

Moreover, the nanobodies are predicted to bind to the functional epitope of Nipah Virus protein which is most relevant for neutralization of the virus.