Pipeline
Description
This workflow uses ESM2 and its log-likelihood scoring function to propose beneficial protein mutations. Starting from a known wild-type sequence and improving its overall log-likelihood (how well the given sequence matches its training data distribution) can yield variants with improved stability, expression, and affinity. These beneficial mutations can be sampled with various algorithms, from greedy search to other balancing between exploration and exploitation. Log-likelihood scores are well-correlated with the aforementioned properties, per the ProteinGym benchmark.
Proteins (23)
id: jade-deer-cedar
EGFR
None
--
True
6.4 kDa
58
id: brisk-yak-ruby
EGFR
None
--
True
6.4 kDa
58
id: quick-moth-plume
EGFR
None
--
True
6.4 kDa
58
id: silver-boar-jade
EGFR
None
--
True
6.4 kDa
58
id: solid-hawk-willow
EGFR
Weak
--
True
6.4 kDa
58
id: vast-quail-cypress
EGFR
None
--
True
6.4 kDa
58
id: bright-otter-iron
EGFR
None
--
True
5.8 kDa
52
id: violet-vole-opal
EGFR
None
--
True
5.4 kDa
49
id: noble-fox-maple
EGFR
None
--
True
5.8 kDa
52
id: rough-falcon-cypress
EGFR
None
--
True
6.5 kDa
58
id: quiet-orca-ruby
EGFR
None
--
True
6.6 kDa
58
id: calm-crane-bronze
EGFR
None
--
True
6.1 kDa
58
id: steady-wolf-stone
EGFR
None
--
True
6.4 kDa
58
id: solid-gecko-cloud
EGFR
None
--
True
6.4 kDa
58
id: swift-dove-opal
EGFR
None
--
True
6.4 kDa
58
id: golden-boar-flint
EGFR
Weak
4.2e-6 M
True
6.7 kDa
58
id: quick-hawk-topaz
EGFR
None
--
True
6.5 kDa
58
id: brisk-swan-frost
EGFR
None
--
True
5.5 kDa
49
id: strong-panda-lava
EGFR
None
--
True
6.5 kDa
58
id: mellow-panther-ember
EGFR
None
--
True
6.6 kDa
58
id: quick-zebra-stone
EGFR
Weak
--
True
5.9 kDa
53