Round 1 campaign targeting PD-L1. Goal: a 55-65 aa de novo minibinder against the GFCC′ face of the human PD-L1 IgV domain — the same face PD-1 and the clinical antibodies atezolizumab/durvalumab occupy — cross-reactive to mouse PD-L1 so a positive hit can move straight into a syngeneic tumor model without a second design cycle.
First order: 24 sequences, generated through RFdiffusion + ProteinMPNN + AF2 initial-guess and cross-checked against BindCraft rather than trusting one engine's output alone — see PD-L1/Note/Engine choice.
Progress
| Stage | Description | Status |
|---|---|---|
| A | Define goal | Done |
| B | Prep target structure — trim to IgV domain only | Done |
| C | Pick epitope and hotspots (GFCC′ face) | Done |
| D | Choose engine — BindCraft + RFdiffusion cross-check | Done |
| E | Generate backbones | Done · PD-L1/Campaign/PDL1-R1/Run/RF-001 |
| F | Design sequences | Done · PD-L1/Campaign/PDL1-R1/Run/MPNN-001 |
| G | In silico filters | Done · PD-L1/Campaign/PDL1-R1/Run/AF2IG-001 — 24 → 6 |
| H | Diversity/developability triage | Done · PD-L1/Campaign/PDL1-R1/Run/ESM-001, PD-L1/Campaign/PDL1-R1/Run/CLUST-001 |
| I | Gene synthesis and expression | Done — 24 ordered, E. coli cytoplasm |
| J | Binding screen | Done — yeast display |
| K | Biophysics (BLI, SEC, nanoDSF) | In progress |
| L | Function (PD-1 blockade) | Partial |
| M | Structure confirmation | 1 |
| N | Round 2 optimization | Planned |
| O | Format conversion | Not started |
Notes
Computed metrics don't predict affinity — passing stage G's BindCraft-style cutoffs (i_pTM, i_pAE, shape complementarity, hydrophobicity, ΔRMSD) means "might bind," not "binds tightly." Actual strength is set at stage K by BLI. In practice, the top-i_pTM design in the shortlist was not the tightest binder measured — see PD-L1/Note/Shortlist rationale for what that miss looked like and why it recurs across binder-design campaigns generally.
Method references
- Reference/Cao 2022 — the target-structure-only design strategy this pipeline descends from
- Reference/RFdiffusion 2023 — stage E, backbone generation
- Reference/ProteinMPNN 2022 — stage F, sequence design
- Reference/Bennett 2023 — stage G, the AF2 initial-guess re-prediction filter
- Reference/BindCraft 2024 — stage D, the cross-check engine
- Reference/ESM-2 2023 — stage H, embedding-based clustering
- Reference/Zak 2015 and Reference/Lee 2017 — the target structure and the atezolizumab epitope benchmark (stages B/C)
Related: PD-L1/Note/Epitope choice · PD-L1/Note/Round 2 plan ·
Protocol/BLI · Protocol/SEC · Protocol/nanoDSF · Protocol/Expression and purification
Sequences and measured values in this dataset are synthetic; the target annotations, structure and references are real.