SMA Research Platform

AI-Designed VHH Nanobodies

AI-GENERATED

20 designs · 9 accepted

Therapeutic-format single-domain antibodies (VHH) designed via the mBER framework using AlphaFold2-Multimer backpropagation to optimize binder sequences against SMA-relevant targets (LIMK2, ROCK2). Every row below is a de-novo designed sequence with computed interface predicted TM-score (ipTM) and per-residue pLDDT confidence. Click any row to open the full sequence with a copy button + AlphaFold2-Multimer complex PDB link.

Source PDBs:sma-research/binders/— full AlphaFold2-Multimer complexes + relaxed binder structures, per-target folders

Total Designs

Accepted (ipTM ≥ 0.7)

Targets

0.791

Best ipTM

0.947

Best pLDDT

0.691

Mean ipTM

LIMK2 · 10 designs

Compound ID	ipTM	pLDDT	Length (aa)	Accepted
LIMK2_4718644_binder-1	0.791	0.936	118	YES
LIMK2_4718644_binder-2	0.789	0.935	118	YES
LIMK2_4718644_binder-0	0.788	0.936	118	YES
LIMK2_4914994_binder-8	0.739	0.941	118	YES
LIMK2_4914994_binder-3	0.714	0.937	118	YES
LIMK2_4914994_binder-5	0.691	0.930	118	YES
LIMK2_4914994_binder-7	0.674	0.919	118	YES
LIMK2_3296102_binder-2	0.672	0.915	118	YES
LIMK2_4914994_binder-4	0.670	0.922	118	YES
LIMK2_4914994_binder-2	0.632	0.919	118	YES

ROCK2 · 10 designs

Compound ID	ipTM	pLDDT	Length (aa)	Accepted
ROCK2_1063592_binder-0	0.750	0.947	118	YES
ROCK2_1626191_binder-2	0.725	0.941	118	YES
ROCK2_1626191_binder-1	0.725	0.939	118	YES
ROCK2_1626191_binder-0	0.722	0.940	118	YES
ROCK2_1063592_binder-3	0.643	0.921	118	YES
ROCK2_3339605_binder-0	0.637	0.933	118	YES
ROCK2_3339605_binder-4	0.630	0.929	118	YES
ROCK2_3339605_binder-6	0.618	0.935	118	YES
ROCK2_3339605_binder-1	0.613	0.935	118	YES
ROCK2_3339605_binder-3	0.601	0.931	118	YES

How does mBER nanobody design work?

mBER (Manifold Binder Engineering and Refinement) designs VHH nanobodies using gradient descent through AlphaFold2-Multimer. Each trajectory: (1) prepares target template structure from PDB; (2) initializes random VHH CDR sequences; (3) backpropagates loss through AF2-Multimer to optimize binder sequence; (4) evaluates interface predicted TM-score (ipTM) and pLDDT; (5) relaxes final structure with OpenMM.

Thresholds:ipTM ≥ 0.7 is “accepted” (mBER paper reports 45% experimental success rate). pLDDT ≥ 0.9 indicates very high structural confidence.

Why VHH format? Single-domain antibodies (nanobodies) are 10× smaller than conventional antibodies (~15 kDa vs 150 kDa), more stable, easier to produce, and can potentially cross the blood-brain barrier when engineered appropriately.

Limitation: These are computational predictions. Experimental binding validation (SPR, ITC, BLI) is required before therapeutic development.