LinkedIn·Sunday, 9 August 2026·17d ago

How convergence separates good designs from lucky ones. (🎥 Designing nanobody Ep8 ) After the multi-seed REC8 screen, we were left with…

Philippe Herman

EVOBodies® for your AI generated and wet-lab validated VHHs in 12 weeks | Minibinders | AMR | CEO IMMUNOCHEM

How convergence separates good designs from lucky ones. (🎥 Designing nanobody Ep8 ) After the multi-seed REC8 screen, we were left with 362 candidates. It is good enough to survive noise, but not good enough to trust with a wet lab budget yet 🙃 For this next cut, let's bring in a second, independent model: Boltz2X, alongside ESM2Fold, and we'll run convergence between the two. We keep the same multi-seed logic as before, no single lucky fold saves a candidate here! 📈 The bar goes up too. Where the first pass only asked for 3 buried epitope residues, this round asks for at least 11, out of 14 possible, to count as positive. ✅ At the end of this step, we're down to 123 candidates. This is the part of the funnel that actually matters for the wet lab : 14,000 to 362 got rid of noise. 362 to 123 is where we start separating good designs from very good ones, based on two independent models agreeing. ➡️ Next post soon : we will bring in an orthogonal validation layer that isn't AI at all. It is molecular dynamics, grounded in actual physics.🌐 Phil. PS: I've launched EVOBody™, a unique platform allowing biotech leaders to develop antibody projects from concept to generation with minimal risk in only 12 weeks, for maximum impact. Apply for access to our program by clicking on 'Schedule a Meeting' under my profile.