After a 50-trajectory mBER run on an A100 SXM4 GPU (~10h compute, ~$11 total), the platform has produced its first publishable quality VHH nanobody designs targeting both ROCK2 and LIMK2 — the two upstream kinases in the SMA-relevant ROCK-LIMK-Cofilin pathway.

Results

ROCK2 Nanobodies — 3 Accepted Binders:

LIMK2 Nanobodies — 1 Accepted Binder:

Overall: 942 total PDB files generated, 40 accepted binders (20 per target), 4 above the 0.7 ipTM publishable threshold.

Why This Matters

The mBER paper (Manifold Bio, bioRxiv 2025) reports a 45% experimental success rate for designs scoring above ipTM 0.7. Our top ROCK2 design (0.7252) and LIMK2 design (0.7142) meet this threshold. pLDDT values above 0.93 indicate very high structural confidence.

These are therapeutic-format single-domain antibodies (VHH nanobodies) — the same modality used in FDA-approved drugs like caplacizumab (Cablivi). VHH nanobodies are:

• 10× smaller than conventional antibodies (~15 kDa)
• Easier to express and more stable
• Can potentially cross the blood-brain barrier when engineered appropriately
• Suitable for targeting both kinase active sites and protein-protein interaction surfaces

Convergent Design

Notable observation: all 3 top ROCK2 binders share the CDR3 motif VRRPTV·DY (different residues at positions 6 and 9), suggesting mBER converged on a specific binding mode. This convergence across independent trajectories is a positive indicator of design stability.

Method

• Framework: mBER VHH binder design (GitHub: manifoldbio/mber-open)
• Compute: A100 SXM4 80GB, ~10h total
• Trajectories: 50 per target, 10 binders per trajectory
• Target PDBs: 2F2U (ROCK2 kinase domain), 4TPT (LIMK2 kinase domain)
• Acceptance threshold: min-ipTM 0.6 (we kept designs ≥0.7 for validation)
• Models: AlphaFold2-Multimer backpropagation + NanoBodyBuilder2 refinement + OpenMM relaxation

Limitations

• Computational predictions only. No experimental binding data yet.
• ROCK2 has a large flat kinase surface — challenging target for de novo binder design
• All 3 accepted LIMK2 candidates come from a single trajectory (4914994) — less design diversity
• Experimental validation via SPR, ITC, or BLI is required before therapeutic development

Next Steps

1. Kinase selectivity: Check cross-reactivity against LIMK1, ROCK1, ABL1, SRC
2. CDR diversity: Run additional trajectories to find diverse binding modes
3. Experimental validation: SPR or BLI binding assay via CRO partnership
4. Humanization: Evaluate framework humanization if binders proceed to preclinical

All designs are viewable at Nanobody Designs with full sequences and structural metadata.