Seeding from round-1 hits:
- Partial diffusion on the confirmed binders, to fine-tune backbones without throwing away a validated global fold
- ProteinMPNN redesign of the interface only, holding the rest of the sequence fixed, for affinity maturation that doesn't disturb the developability profile already confirmed at stage I
- Point-mutation scan via display DMS — cheap, high-throughput, and catches single-residue affinity gains a redesign round would miss
- Disulfide stapling on the least stable hits to raise Tm
- Trimerization for avidity, for any hit whose monomer KD plateaus rather than improves with further maturation
Prescription by failure mode, from the round-1 assay data:
- Binds but weak (BLI KD in the hundreds of nM): affinity maturation — interface-only MPNN redesign, expand the contact patch toward adjacent hotspots, or run the display DMS scan before committing to a redesign
- Aggregates / poor SEC monomer fraction: lower computed surface hydrophobicity in the next MPNN pass and re-check for exposed hydrophobic runs — see PD-L1/Note/Expression results for which round-1 hits triggered this
- Doesn't express or stays insoluble: try periplasmic or oxidizing expression before writing off the sequence — see Protocol/Expression and purification
- Binds nonspecifically in controls: recharge the surface (fewer exposed hydrophobic/aromatic patches outside the interface) and add a second orthogonal negative control to the panel
- Binds and folds cleanly but fails the blockade assay: wrong face or wrong orientation, not a biophysics problem — revisit PD-L1/Note/Epitope choice before touching the sequence again