Designs were guided by multiple known VHH sequences or structures, and included ESM-3 architecture, MPNN/RFDiffusion of CDRs, AntiFold, and a BindCraft-inspired nanobody ddG method. Top ranked sequences were chosen diversely using predictive metrics like ddG, log-probabilities from models like NanoBERT, and other methods, intended to create a high-variance and low-bias experimental dataset.
id: bright-swan-orchid

EGFR
Strong
1.4e-8 M
True
14.0 kDa
127
id: gentle-mole-willow

EGFR
Strong
1.4e-8 M
True
14.5 kDa
130
id: solid-eagle-ash

EGFR
Strong
1.5e-8 M
True
13.1 kDa
123
id: quiet-yak-ice

EGFR
Strong
5.7e-8 M
True
13.3 kDa
122
id: violet-raven-iron

EGFR
Strong
2.1e-8 M
True
12.6 kDa
119
id: bright-kiwi-lotus

EGFR
Medium
4.4e-8 M
True
14.4 kDa
132
id: crimson-toad-onyx

EGFR
Medium
5.8e-8 M
True
13.4 kDa
124
id: crimson-falcon-pine

EGFR
Medium
1.1e-7 M
True
13.3 kDa
122
id: golden-panda-cloud

EGFR
Medium
1.5e-7 M
True
13.6 kDa
122
id: jade-ox-dust

EGFR
Medium
5.1e-7 M
True
13.3 kDa
122
id: noble-yak-pine

EGFR
Medium
1.8e-7 M
True
13.5 kDa
122
id: rough-yak-clay

EGFR
Medium
2.5e-7 M
True
13.1 kDa
122
id: young-seal-lotus

EGFR
None
--
True
12.9 kDa
122
id: vast-ibis-stone

EGFR
None
--
True
13.1 kDa
123
id: violet-deer-leaf

EGFR
None
--
True
13.3 kDa
122
id: dark-mole-lotus

EGFR
None
--
True
12.8 kDa
122
id: strong-boar-rose

EGFR
None
--
True
13.3 kDa
124
id: green-tiger-reed

EGFR
None
--
True
12.9 kDa
124
id: azure-eagle-thorn

EGFR
None
--
True
12.8 kDa
121
id: small-lynx-moss

EGFR
None
--
True
13.1 kDa
122
id: pale-vole-willow

EGFR
None
--
True
13.4 kDa
122